This is a re-publishing of a blog post I originally wrote for work, but wanted on my own blog as well. AI is everywhere, and its impressive claims are leading to rapid adoption. At this stage, I’d qualify it as charismatic technology—something that under-delivers on what it promises, but promises so much that the industry still leverages it because we believe it will eventually deliver on these claims. This is a known pattern. In this post, I’ll use the example of automation deployments to go over known patterns and risks in order to provide you with a list of questions to ask about potential AI solutions. I’ll first cover a short list of base assumptions, and then borrow from scholars of cognitive systems engineering and resilience engineering to list said criteria. At the core of it is the idea that when we say we want humans in the loop, it really matters where in the loop they are. My base assumptions The first thing I’m going to say is that we currently do not have Artificial General Intelligence (AGI). I don’t care whether we have it in 2 years or 40 years or never; if I’m looking to deploy a tool (or an agent) that is supposed to do stuff to my production environments, it has to be able to do it now. I am not looking to be impressed, I am looking to make my life and the system better. Another mechanism I want you to keep in mind is something called the context gap. In a nutshell, any model or automation is constructed from a narrow definition of a controlled environment, which can expand as it gains autonomy, but remains limited. By comparison, people in a system start from a broad situation and narrow definitions down and add constraints to make problem-solving tractable. One side starts from a narrow context, and one starts from a wide one—so in practice, with humans and machines, you end up seeing a type of teamwork where one constantly updates the other: The optimal solution of a model is not an optimal solution of a problem unless the model is a perfect representation of the problem, which it never is.  — Ackoff (1979, p. 97) Because of that mindset, I will disregard all arguments of “it’s coming soon” and “it’s getting better real fast” and instead frame what current LLM solutions are shaped like: tools and automation. As it turns out, there are lots of studies about ergonomics, tool design, collaborative design, where semi-autonomous components fit into sociotechnical systems, and how they tend to fail. Additionally, I’ll borrow from the framing used by people who study joint cognitive systems: rather than looking only at the abilities of what a single person or tool can do, we’re going to look at the overall performance of the joint system. This is important because if you have a tool that is built to be operated like an autonomous agent, you can get weird results in your integration. You’re essentially building an interface for the wrong kind of component—like using a joystick to ride a bicycle. This lens will assist us in establishing general criteria about where the problems will likely be without having to test for every single one and evaluate them on benchmarks against each other. Questions you'll want to ask The following list of questions is meant to act as reminders—abstracting away all the theory from research papers you’d need to read—to let you think through some of the important stuff your teams should track, whether they are engineers using code generation, SREs using AIOps, or managers and execs making the call to adopt new tooling. Are you better even after the tool is taken away? An interesting warning comes from studying how LLMs function as learning aides. The researchers found that people who trained using LLMs tended to fail tests more when the LLMs were taken away compared to people who never studied with them, except if the prompts were specifically (and successfully) designed to help people learn. Likewise, it’s been known for decades that when automation handles standard challenges, the operators expected to take over when they reach their limits end up worse off and generally require more training to keep the overall system performant. While people can feel like they’re getting better and more productive with tool assistance, it doesn’t necessarily follow that they are learning or improving. Over time, there’s a serious risk that your overall system’s performance will be limited to what the automation can do—because without proper design, people keeping the automation in check will gradually lose the skills they had developed prior. Are you augmenting the person or the computer? Traditionally successful tools tend to work on the principle that they improve the physical or mental abilities of their operator: search tools let you go through more data than you could on your own and shift demands to external memory, a bicycle more effectively transmits force for locomotion, a blind spot alert on your car can extend your ability to pay attention to your surroundings, and so on. Automation that augments users therefore tends to be easier to direct, and sort of extends the person’s abilities, rather than acting based on preset goals and framing. Automation that augments a machine tends to broaden the device’s scope and control by leveraging some known effects of their environment and successfully hiding them away. For software folks, an autoscaling controller is a good example of the latter. Neither is fundamentally better nor worse than the other—but you should figure out what kind of automation you’re getting, because they fail differently. Augmenting the user implies that they can tackle a broader variety of challenges effectively. Augmenting the computers tends to mean that when the component reaches its limits, the challenges are worse for the operator. Is it turning you into a monitor rather than helping build an understanding? If your job is to look at the tool go and then say whether it was doing a good or bad job (and maybe take over if it does a bad job), you’re going to have problems. It has long been known that people adapt to their tools, and automation can create complacency. Self-driving cars that generally self-drive themselves well but still require a monitor are not effectively monitored. Instead, having AI that supports people or adds perspectives to the work an operator is already doing tends to yield better long-term results than patterns where the human learns to mostly delegate and focus elsewhere. (As a side note, this is why I tend to dislike incident summarizers. Don’t make it so people stop trying to piece together what happened! Instead, I prefer seeing tools that look at your summaries to remind you of items you may have forgotten, or that look for linguistic cues that point to biases or reductive points of view.) Does it pigeonhole what you can look at? When evaluating a tool, you should ask questions about where the automation lands: Does it let you look at the world more effectively? Does it tell you where to look in the world? Does it force you to look somewhere specific? Does it tell you to do something specific? Does it force you to do something? This is a bit of a hybrid between “Does it extend you?” and “Is it turning you into a monitor?” The five questions above let you figure that out. As the tool becomes a source of assertions or constraints (rather than a source of information and options), the operator becomes someone who interacts with the world from inside the tool rather than someone who interacts with the world with the tool’s help. The tool stops being a tool and becomes a representation of the whole system, which means whatever limitations and internal constraints it has are then transmitted to your users. Is it a built-in distraction? People tend to do multiple tasks over many contexts. Some automated systems are built with alarms or alerts that require stealing someone’s focus, and unless they truly are the most critical thing their users could give attention to, they are going to be an annoyance that can lower the effectiveness of the overall system. What perspectives does it bake in? Tools tend to embody a given perspective. For example, AIOps tools that are built to find a root cause will likely carry the conceptual framework behind root causes in their design. More subtly, these perspectives are sometimes hidden in the type of data you get: if your AIOps agent can only see alerts, your telemetry data, and maybe your code, it will rarely be a source of suggestions on how to improve your workflows because that isn’t part of its world. In roles that are inherently about pulling context from many disconnected sources, how on earth is automation going to make the right decisions? And moreover, who’s accountable for when it makes a poor decision on incomplete data? Surely not the buyer who installed it! This is also one of the many ways in which automation can reinforce biases—not just based on what is in its training data, but also based on its own structure and what inputs were considered most important at design time. The tool can itself become a keyhole through which your conclusions are guided. Is it going to become a hero? A common trope in incident response is heroes—the few people who know everything inside and out, and who end up being necessary bottlenecks to all emergencies. They can’t go away for vacation, they’re too busy to train others, they develop blind spots that nobody can fix, and they can’t be replaced. To avoid this, you have to maintain a continuous awareness of who knows what, and crosstrain each other to always have enough redundancy. If you have a team of multiple engineers and you add AI to it, having it do all of the tasks of a specific kind means it becomes a de facto hero to your team. If that’s okay, be aware that any outages or dysfunction in the AI agent would likely have no practical workaround. You will essentially have offshored part of your ops. Do you need it to be perfect? What a thing promises to be is never what it is—otherwise AWS would be enough, and Kubernetes would be enough, and JIRA would be enough, and the software would work fine with no one needing to fix things. That just doesn’t happen. Ever. Even if it’s really, really good, it’s gonna have outages and surprises, and it’ll mess up here and there, no matter what it is. We aren’t building an omnipotent computer god, we’re building imperfect software. You’ll want to seriously consider whether the tradeoffs you’d make in terms of quality and cost are worth it, and this is going to be a case-by-case basis. Just be careful not to fix the problem by adding a human in the loop that acts as a monitor! Is it doing the whole job or a fraction of it? We don’t notice major parts of our own jobs because they feel natural. A classic pattern here is one of AIs getting better at diagnosing patients, except the benchmarks are usually run on a patient chart where most of the relevant observations have already been made by someone else. Similarly, we often see AI pass a test with flying colors while it still can’t be productive at the job the test represents. People in general have adopted a model of cognition based on information processing that’s very similar to how computers work (get data in, think, output stuff, rinse and repeat), but for decades, there have been multiple disciplines that looked harder at situated work and cognition, moving past that model. Key patterns of cognition are not just in the mind, but are also embedded in the environment and in the interactions we have with each other. Be wary of acquiring a solution that solves what you think the problem is rather than what it actually is. We routinely show we don’t accurately know the latter. What if we have more than one? You probably know how straightforward it can be to write a toy project on your own, with full control of every refactor. You probably also know how this stops being true as your team grows. As it stands today, a lot of AI agents are built within a snapshot of the current world: one or few AI tools added to teams that are mostly made up of people. By analogy, this would be like everyone selling you a computer assuming it were the first and only electronic device inside your household. Problems arise when you go beyond these assumptions: maybe AI that writes code has to go through a code review process, but what if that code review is done by another unrelated AI agent? What happens when you get to operations and common mode failures impact components from various teams that all have agents empowered to go fix things to the best of their ability with the available data? Are they going to clash with people, or even with each other? Humans also have that ability and tend to solve it via processes and procedures, explicit coordination, announcing what they’ll do before they do it, and calling upon each other when they need help. Will multiple agents require something equivalent, and if so, do you have it in place? How do they cope with limited context? Some changes that cause issues might be safe to roll back, some not (maybe they include database migrations, maybe it is better to be down than corrupting data), and some may contain changes that rolling back wouldn’t fix (maybe the workload is controlled by one or more feature flags). Knowing what to do in these situations can sometimes be understood from code or release notes, but some situations can require different workflows involving broader parts of the organization. A risk of automation without context is that if you have situations where waiting or doing little is the best option, then you’ll need to either have automation that requires input to act, or a set of actions to quickly disable multiple types of automation as fast as possible. Many of these may exist at the same time, and it becomes the operators’ jobs to not only maintain their own context, but also maintain a mental model of the context each of these pieces of automation has access to. The fancier your agents, the fancier your operators’ understanding and abilities must be to properly orchestrate them. The more surprising your landscape is, the harder it can become to manage with semi-autonomous elements roaming around. After an outage or incident, who does the learning and who does the fixing? One way to track accountability in a system is to figure out who ends up having to learn lessons and change how things are done. It’s not always the same people or teams, and generally, learning will happen whether you want it or not. This is more of a rhetorical question right now, because I expect that in most cases, when things go wrong, whoever is expected to monitor the AI tool is going to have to steer it in a better direction and fix it (if they can); if it can’t be fixed, then the expectation will be that the automation, as a tool, will be used more judiciously in the future. In a nutshell, if the expectation is that your engineers are going to be doing the learning and tweaking, your AI isn’t an independent agent—it’s a tool that cosplays as an independent agent. Do what you will—just be mindful All in all, none of the above questions flat out say you should not use AI, nor where exactly in the loop you should put people. The key point is that you should ask that question and be aware that just adding whatever to your system is not going to substitute workers away. It will, instead, transform work and create new patterns and weaknesses. Some of these patterns are known and well-studied. We don’t have to go rushing to rediscover them all through failures as if we were the first to ever automate something. If AI ever gets so good and so smart that it’s better than all your engineers, it won’t make a difference whether you adopt it only once it’s good. In the meanwhile, these things do matter and have real impacts, so please design your systems responsibly. If you’re interested to know more about the theoretical elements underpinning this post, the following references—on top of whatever was already linked in the text—might be of interest: Books: Joint Cognitive Systems: Foundations of Cognitive Systems Engineering by Erik Hollnagel Joint Cognitive Systems: Patterns in Cognitive Systems Engineering by David D. Woods Cognition in the Wild by Edwin Hutchins Behind Human Error by David D. Woods, Sydney Dekker, Richard Cook, Leila Johannesen, Nadine Sarter Papers: Ironies of Automation by Lisanne Bainbridge The French-Speaking Ergonomists’ Approach to Work Activity by Daniellou How in the World Did We Ever Get into That Mode? Mode Error and Awareness in Supervisory Control by Nadine Sarter Can We Ever Escape from Data Overload? A Cognitive Systems Diagnosis by David D. Woods Ten Challenges for Making Automation a “Team Player” in Joint Human-Agent Activity by Gary Klein and David D. Woods MABA-MABA or Abracadabra? Progress on Human–Automation Co-ordination by Sidney Dekker Managing the Hidden Costs of Coordination by Laura Maguire Designing for Expertise by David D. Woods The Impact of Generative AI on Critical Thinking by Lee et al.

I like to think that I write code deliberately. I’m an admittedly slow developer, and I want to believe I do so on purpose. I want to know as much as I can about the context of what it is that I'm automating. I also use a limited set of tools. I used old computers for a long time, both out of an environmental mindset, but also because a slower computer quickly makes it obvious when something scales poorly.1 The idea is to seek friction, and harness it as an early signal that whatever I’m doing may need to be tweaked, readjusted. I find this friction, and even frustration in general to also be useful around learning approaches.2 In opposition to the way I'd like to do things, everything about the tech industry is oriented towards elevated productivity, accelerated growth, and "easy" solutions to whole families of problems. I feel that maybe we should teach people to program the way they teach martial arts, like only in the most desperate situations when all else failed should you resort to automating something. I don’t quite know if I’m just old and grumpy, seeing industry trends fly by me at a pace I don’t follow, or whether there’s really something to it, but I thought I’d take a walk through a set of ideas and concepts that motivate my stance. This blog post has a lot of ground to cover. I'll first start with some fundamental properties of systems and how overload propagates through various bottlenecks. Then I'll go over some high-level pressures that are shared by most organizations and force trade-offs down their structure. These two aspects—load propagation and pervasive trade-offs—create the need for compensatory actions, of which we'll discuss some limits. This, finally, will be tied back to friction and ways to listen to it, because it's one of the things that underpins adaptation and keeps systems running. Optimizing While Leaving Pressures in Place Optimizing a frictional path without revising the system’s conditions and pressures tends to not actually improve the system. Instead, what you’re likely to do is surface brittleness in all the areas that are now exposed to the new system demands. Whether a bottleneck was invisible or well monitored, and regardless of scale, it offered an implicit form of protection that was likely taken for granted. For a small scale example, imagine you run a small bit of software on a server, talking to a database. If you suddenly get a lot of visits, simply autoscaling the web front-end will likely leave the database unprotected and sensitive to tipping over (well, usually after having grown the connection pool, raised the connection limit, vertically scaled the servers, and so on). None of this will let you serve heavy traffic at a reasonable price until you rework your caching and data distribution strategy. Building for orders of magnitude more traffic than usual requires changing some fundamental aspects of your solution. Similar patterns can be seen at a larger scale. An interesting case was the Clarkesworld magazine; as LLMs made it possible to produce slop at a faster rate than previously normal, an inherent bottleneck in authorship ("writing a book takes significant time and effort") was removed, leading to so much garbage that the magazine had to stop taking in submissions. They eventually ended up bearing the cost of creating a sort of imperfect queuing "spam filter" for submissions in order to accept them again. They don't necessarily publish more stories than before, they still aim to publish the good human-written stuff, there's just more costly garbage flowing through the system.3 A similar case to look for is how doctors in the US started using generative AI to fight insurance claim denials. Of course, insurers are now expected to adopt the same technology to counteract this effect. A general issue at play here is that the private insurance system's objectives and priorities are in conflict with those of the doctors and patients. Without realigning them, most of what we can expect is an increase in costs and technological means to get the same results out of it. People who don’t or can’t use the new tools are going to be left behind. The optimization's benefit is temporary, limited, and ultimately lost in the overall system, which has grown more complex and possibly less accessible.4 I think LLMs are top of mind for people because they feel like a shift in how you automate. The common perspective is that machines are good at repetitive, predictable, mechanical tasks, and that solutions always suffered when it came to the fuzzy, unpredictable, and changing human-adjacent elements. LLMs look exactly the opposite of that: the computers can't do math very well anymore, but they seem to hold conversations and read intent much better. They therefore look like a huge opportunity to automate more of the human element and optimize it away, following well-established pressures and patterns. Alternatively, they seemingly increase the potential for new tools that could be created and support people in areas where none existed before. The issues I'm discussing here clearly apply to AI, Machine Learning, and particularly LLMs. But they also are not specific to them. People who love the solution more than they appreciate the problem risk delivering clumsy integrations that aren’t really fit for purpose. This is why it feels like companies are wedging more AI in our face; that's what the investors wanted in order to signal innovativeness, or because the engineers really wanted to build cool shit, rather than solving the problems the users wanted or needed solved. The challenges around automation were always there from their earliest days and keep being in play now. They remain similar without regards to the type of automation or optimization being put in place, particularly if the system around them does not reorganize itself. The canonical example here is what happens when an organization looms so large that people can't understand what is going on. The standard playbook around this is to start driving purely by metrics, which end up compressing away rich phenomena. Doing so faster, whether it is by gathering more data (even if we already had too much) or by summarizing harder via a LLM likely won't help run things better. Summaries, like metrics, are lossy compression. They're also not that different from management by PowerPoint slides, which we've seen cause problems in the space program, as highlighted by the Columbia report: As information gets passed up an organization hierarchy, from people who do analysis to mid-level managers to high-level leadership, key explanations and supporting information is filtered out. In this context, it is easy to understand how a senior manager might read this PowerPoint slide and not realize that it addresses a life-threatening situation. At many points during its investigation, the Board was surprised to receive similar presentation slides from NASA officials in place of technical reports. The Board views the endemic use of PowerPoint briefing slides instead of technical papers as an illustration of the problematic methods of technical communication at NASA. There is no reason to think that overly aggressive summarization via PowerPoint, LLM, or metrics would not all end similarly. If your decision-making layer cannot deal with the amount of information required to centrally make informed decisions, there may be a point where the solution is to change the system's structure (and decentralize, which has its own pitfalls) rather than to optimize the existing paths without question.5 Every actor, component, or communication channel in a system has inherent limits. Any part that suddenly becomes faster or more productive without feedback shifts greater burdens onto other parts. These other parts must adapt, adjust, pass on the cost, or stop meeting expectations. Eliminating friction from one part of the system sometimes just shifts it around. System problems tend to remain system problems regardless of how much you optimize isolated portions of them. Pressures and Propagation How can we know what is worth optimizing, and what is changing at a more structural level?6 It helps to have an idea of where the pressures that create goal conflicts might come from, since they eventually lead to adaptations. Systems tend to continually be stretched to the limit of their capacity, and any improvement is instantly leveraged to accelerate the pace of existing activities. This is usually where online people say things like "the root cause is capitalism"7—you shouldn't expect local solutions to fix systemic problems in the long term. The moment other players dynamically reduce their margins of maneuver to gain efficiency, you become relatively less competitive. You can think of how we could all formally prove software to be safe before shipping it, but instead we’ll compromise by using less formal methods like type analysis, tests, or feature flags to deliver acceptable products at much lower costs—both financial and cognitive. Be late to the market and you suffer, so there's a constant drive to ship faster and course-correct often. People more hopeful or trusting of a system try to create and apply counteracting forces to maintain safe operating margins. This tends to be done through changing incentives, creating regulatory bodies, and implementing better control and reporting mechanisms. This is often the approach you'll see taken around the nuclear industry, the FAA and the aviation industry, and so on. However, there are also known patterns (such as regulatory capture) that tend to erode these mechanisms, and even within each of these industries, surprises and adaptations are still a regular occurrence. Ultimately, the effects of any technological change are rather unpredictable. Designing for systems where experts operate demands constantly revisiting and iterating. The concepts we define to govern systems create their own indifference to other important perspectives, and data-driven approaches carry the risk of "bias laundering" mechanisms that repeat and amplify existing flaws in the system. Other less predictable effects can happen. Adopting objectively more accurate algorithms can create monocultures in decision-making, which can interact such that the overall system efficiency can go down compared to more diverse environments—even in the absence of disruption. Basically, the need for increased automation isn't likely to "normalize" a system and make it more predictable. It tends to just create new types of surprises in a way that does not remove the need for adaptation nor shift pressures; it only transforms them and makes them dynamic. Robust Yet Fragile Embedded deeply in our view of systems is an assumption that things are stable until they are disrupted. It’s possibly where ideas like “root cause” gain their charisma: identify the one triggering disruptor (or its underlying mechanism) and then the system will be stable again. It’s conceptually a bit Newtonian in that if no force is applied, nothing will change. A more ecological stance would instead assume that any perceived stability (while maintaining function) requires ongoing dynamic adjustments. The system is always decaying, transforming, interacting, changing. Stop interfering with it and it will eventually reach stability (without maintaining function) by breaking down or failing. If the pressures are constant and shifting as well as the counteracting mechanisms, we can assume that evolution and adaptation are required to deal with this dynamism. Over time, we should expect that the system instead evolves into a shape that fits its burdens while driven by scarcity and efficiency. A risk in play here is that an ecosystem's pressures make it rational and necessary for all actors to optimize when they’re each other’s focal point—rather than some environmental condition. The more aggressively it is done, the more aggressively it is needed by others to stay in the game. Robust yet fragile is the nature of systems that are well optimized for their main use cases and competitive within their environment, but which become easily upended by pressures applied from unexpected angles (that are therefore unprotected, since resources were used elsewhere instead). Good examples of this are Just-In-Time supply chains being far more efficient than traditional ones, but being far easier to disrupt in times of disasters or pandemics. Most buffers in the supply chain (such as stock held in warehouses) had been replaced by more agile and effective production and delivery mechanisms. Particularly, the economic benefits (in stable times) and the need for competitiveness have made it tricky for many businesses not to rely on them. The issue with optimizations driven from systemic pressures is that as you look at trimming the costs of keeping a subsystem going in times of stability, you may notice decent amounts of slack capacity that you could get rid of or drive harder in order to be more competitive in your ecosystem. That’s often resources that resilience efforts draw on to keep adapting and evolving. Another form of rationalization in systems is one where rather than cutting "excess", the adoption and expansion of (software) platforms are used to drive economies of scale. Standardization and uniformization of patterns, methods, and processes is a good way to get more bang for your buck on an investment, to do more with less. Any such platform is going to have some things it gives its users for cheap, and some things that become otherwise challenging to do.8 Friction felt here can both be caused by going against the platform's optimal use cases or by the platform not properly supporting some use cases—it's a signal worth listening to. In fact, we can more or less assume that friction is coming from everywhere because it's connected to these pressures. They just happen to be pervasive, at every layer of abstraction. If we had infinite time, infinite resources, or infinite capacity, we'd never need to optimize a thing. Compensatory Adaptive Mechanisms Successfully navigating these pressures is essentially drawing from concepts such as graceful extensibility and sustained adaptability. In a nutshell, we're looking to know how systems stretch themselves to deal with disruptions and surprises in a context of finite resources, and also how a system manages and regulates its own abilities to do that on an ongoing basis. Remember that every actor or component of a system has inherent limits. This is also true of our ability to know what is going on, something known as local rationality. This means that even if we're really hoping we could intervene from the system level first and avoid the (sometimes deceptively ineffective) local optimizations, it will regardless be attempted through local efforts. Knowing and detecting the friction behind it is useful for whoever wants the broader systematic view to act earlier, but large portions of the system are going to remain dynamic and co-evolving from locally felt pains and friction. Local rationality impacts everyone, even the most confident of system thinkers. Friction shifts are unavoidable, so it's useful to also know of the ways in which they show up. Unfortunately, these shifts generally remain unseen from afar, because compensatory mechanisms and adaptation patterns hide them.9. So instead, it's more practical to find how to spot the compensatory patterns themselves. One of the well-known mechanisms is the Efficiency–thoroughness trade-off (ETTO) principle, which states that since time and resources are limited, one has to trade-off efficiency and thoroughness to accomplish a task. Basically, if there's more work to do than there's capacity to do it, either you maintain thoroughness and the work accumulates or gets dropped, or you do work less thoroughly, possibly cut corners, accuracy, or you have to be less careful and keep going as fast as required. This is also one of the patterns feeding concepts such as "deviance" (often used in normalization of deviance, although the term alone points to any variation relative to norms), where procedures and rules defining safe work start being modified or bent unofficially, until covert work patterns grow a gap between the work as it is specified and how it is practiced.10 Of course, another path is one of innovation, which can mean some reorganization or restructuring. We happen to be in tech, so we tend to prefer to increase capacity by using new technology. New technology is rarely neutral and never isolated. It disturbs established patterns—often on purpose, but sometimes in unexpected ways—can require a complex support system, and for everyone to adjust around it to maintain the proper operational context. Adding to this, if automation is clumsy enough, it won’t be used to its full potential to avoid distracting or burdening practitioners using it to do their work. The ongoing adaptations and trade-offs create potential risks and needs for reciprocity to anticipate and respond to new contingencies. You basically need people who know the system, how it works, understand what is normal or abnormal, and how to work around its flaws. They are usually those who have the capacity to detect any sort of "creaking" in local parts of the system, who harness the friction and can then do some adjusting, mustering and creating slack to provide the margin to absorb surprises. They are compensating for weaknesses as they appear by providing adaptive capacity. Some organizations may enjoy these benefits without fixing anything else by burning out employees and churning through workers, using them as a kind of human buffer for systemic stressors. This can sustain them for a while, but may eventually reach its limits. Even without any sort of willful abuse, pressures lead a system to try to fully use or optimize away the spare capacity within. This can eventually exhaust the compensatory mechanisms it needs to function, leading to something called "decompensation". Decompensation Compensatory mechanisms are often called on so gradually that your average observer wouldn't even know it's taking place. Systems (or organisms) that appear absolutely healthy one day collapse, and we discover they were overextended for a long while. Let's look at congestive heart failure as an example.11 Effects of heart damage accumulate gradually over the years—partly just by aging—and can be offset by compensatory mechanisms in the human body. As the heart becomes weaker and pumps less blood with each beat, adjustments manage to keep the overall flow constant over time. This can be done by increasing the heart rate using complex neural and hormonal signaling. Other processes can be added to this: kidneys faced with lower blood pressure and flow can reduce how much urine they create to keep more fluid in the circulatory system, which increases cardiac filling pressure, which stretches the heart further before each beat, which adds to the stroke volume. Multiple pathways of this kind exist through the body, and they can maintain or optimize cardiac performance. However, each of these compensatory mechanisms has less desirable consequences. The heart remains damaged and they offset it, but the organism remains unable to generate greater cardiac output such as would be required during exercise. You would therefore see "normal" cardiac performance at rest, with little ability to deal with increased demand. If the damage is gradual enough, the organism will adjust its behavior to maintain compensation: you will walk slower, take breaks while climbing stairs, and will just generally avoid situations that strain your body. This may be done without even awareness of the decreased capacity of the system, and we may even resist acknowledging that we ever slowed down. Decompensation happens when all the compensatory mechanisms no longer prevent a downward spiral. If the heart can't maintain its output anymore, other organs (most often the kidneys) start failing. A failing organ can't overextend itself to help the heart; what was a stable negative feedback loop becomes a positive feedback loop, which quickly leads to collapse and death. Someone with a compensated congestive heart failure appears well and stable. They have gradually adjusted their habits to cope with their limited capacity as their heart weakened through life. However, looking well and healthy can hide how precarious of a position the organism is in. Someone in their late sixties skipping their heart medication for a few days or adopting a saltier diet could be enough to tip the scales into decompensation. Decompensation usually doesn’t happen because compensation mechanisms fail, but because their range is exhausted. A system that is compensating looks fine until it doesn’t. That's when failures may cascade and major breakdowns occur. This applies to all sorts of systems, biological as well as sociotechnical. A common example seen in the tech industry is one where overburdened teams continuously pull small miracles and fight fires, keeping things working through major efforts. The teams are stretched thin, nobody's been on vacation for a while, and hiring is difficult because nobody wants to jump into that sort of place. All you need is one extra incident, one person falling ill or quitting, needing to add one extra feature (which nobody has bandwidth to work on), and the whole thing falls apart. But even within purely technical subsystems, automation reaching its limits often shows up a bit like decompensation when it hands control back to a human operator who doesn't have the capacity to deal with what is going on (one of the many things pointed out by the classic text on the Ironies of Automation). Think of an autopilot that disengages once it reached the limit of what it can do to stabilize a plane in hazardous conditions. Or of a cluster autoscaler that can no longer schedule more containers or hosts and starts crowding them until performance collapses, queues fill up, and the whole application becomes unresponsive. Eventually, things spin out into a much bigger emergency than you'd have expected as everything appeared fine. There might have been subtle clues—too subtle to be picked up without knowing where to look—which shouldn't distract from their importance. Friction usually involves some of these indicators. Seeking the Friction Going back to friction being useful feedback, the question I want to ask is: how can we keep listening? The most effective actions are systemic, but the friction patterns are often local. If we detect the friction, papering over it via optimization or brute-force necessarily keeps it local, and potentially ineffective. We need to do the more complex work of turning friction into a system-level feedback signal for it to have better chances of success and sustainability. We can't cover all the clues, but surfacing key ones can be critical for the system to anticipate surprises and foster broader adaptive responses. When we see inappropriate outcomes of a system, we should be led to wonder what about its structure makes it a normal output. What are the externalities others suffer as a consequence of the system's strengths and weaknesses? This is a big question that feels out of reach for most, and not necessarily practical for everyday life. But it’s an important one as we repeatedly make daily decisions around trading off “working a bit faster” against the impacts of the tools we adopt, whether they are environmental, philosophical, or sociopolitical. Closer to our daily work as developers, when we see code that’s a bit messy and hard to understand, we either slow down to create and repair that understanding, or patch it up with local information and move on. When we do this with a tool that manages the information for us, are we in a situation where we accelerate ourselves by providing better framing and structure, or one where we just get where we want without acknowledging the friction?12 If it's the latter, what are the effects of ignoring the friction? Are we creating technical debt that can’t be managed without the tools? Are we risking increasingly not reorganizing the system when it creaks, and only waiting to see obvious breaks to know it needs attention? In fact, how would you even become good at knowing what creaking sounds like if you just always slam through the hurdles? Recognizing these patterns is a skill, and it tends to require knowing what “normal” feels like such that you can detect what is not there when you start deviating.13 If you use a bot for code reviews, ask yourself whether it is replacing people reviewing and eroding the process. Is it providing a backstop? Are there things it can't know about that you think are important? Is it palliating already missing support? Are the additional code changes dictated by review comments worth more than the acts of reviewing and discussing the code? Do you get a different result if the bot only reviews code that someone else already reviewed to add more coverage, rather than implicitly making it easier to ignore reviews and go fast? Work that takes time is a form of friction, and it's therefore tempting to seek ways to make it go faster. Before optimizing it away, ask yourself whether it might have outputs other than its main outputs. Maybe you’re fixing a broken process for an overextended team. Maybe you’re eroding annoying but surprisingly important opportunities for teams to learn, synchronize, share, or reflect on their practices without making room for a replacement. When you're reworking a portion of a system to make it more automatable, ask whether any of the facilitating and structuring steps you're putting in place could also benefit people directly. I recall hearing a customer who said “We are now documenting things in human-readable text so AI can make use of it”—an investment that clearly could have been worth it for people too. Use the change of perspective as an opportunity to surface elements hidden in the broader context and ecosystem, and on which people rely implicitly. I've been disappointed by proposals of turning LLMs into incident reviewers; I'd rather see them becoming analysis second-guessers: maybe they can point out agentive language leading to bias, elements that sound counterfactual, highlights elements that appear blameful to create blame awareness? If you make the decision to automate, still ask the questions and seek the friction. Systems adjust themselves and activate their adaptive capacity based on the type of challenges they face. Highlight friction. It’s useful, and it would be a waste to ignore it. Thanks to Jordan Goodnough, Alan Kraft, and Laura Nolan for reviewing this text. 1: I’m forced to refresh my work equipment more often now because new software appears to hunger for newer hardware at an accelerating pace. 2: As a side note, I'd like to call out the difference between friction, where you feel resistance and that your progression is not as expected based on experience, and one of pain, where you're just making no progress at all and having a plain old bad time. I'd put "pain" in a category where you might feel more helpless, or do useless work just because that's how people first gained the experience without any good reason for it to still be learned the same today. Under this casual definition, friction is the unfamiliar feeling when getting used to your tools and seeking better ways of wielding them, and pain is injuring yourself because the tools have poor ergonomic properties. 3: the same problem can be felt in online book retail, where spammers started hijacking the names of established authors with fake books. The cost of managing this is left to authors—and even myself, having published mostly about Erlang stuff, have had at least two fake books published under my name in the last couple years. 4: In Energy and Equity, Ivan Illich proposes that societies built on high-speed motorized transportation create a "radical monopoly," basically stating that as the society grows around cars and scales its distances proportionally to time spent traveling, living without affording a car and its upkeep becomes harder and harder. This raises the bar of participation in such environments, and it's easy to imagine a parallel within other sociotechnical systems. 5: AI is charismatic technology. It is tempting to think of it as the one optimization that can make decisions such that the overall system remains unchanged while its outputs improve. Its role as fantasized by science fiction is one of an industrial supply chain built to produce constantly good decisions. This does not reduce its potential for surprise or risk. Machine-as-human-replacement is most often misguided. I don't believe we're anywhere that point, and I don't think it's quite necessary to make an argument about it. 6: Because structural changes often require a lot more time and effort than local optimizations, you sometimes need to carry both types of interventions at the same time: a piecemeal local optimization to "extend the runway", and broader interventions to change the conditions of the system. A common problem for sustainability is to assume that extending the runway forever is both possible and sufficient, and never follow up with broader acts. 7: While capitalism has a keen ability to drive constraints of this kind, scarcity constraints are fairly universal. For example, Sonja D. Schmid, in Producing Power illustrates that some of the contributing factors that encouraged the widespread use of the RBMK reactor design in the USSR—the same design used in Chernobyl—were that its manufacturing was more easily distributed over broad geographic areas and sourced from local materials which could avoid the planned system's inefficiencies, and therefore meet electrification objectives in ways that couldn't be done with competing (and safer) reactor designs. Additionally, competing designs often needed centralized manufacturing of parts that could then not be shipped through communist USSR without having to increase the dimensions of some existing train tunnels, forcing upgrades to its rail network to open power plants. An entirely unrelated example is that a beehive's honeycomb structure optimizes for using the least material to create a lattice of cells within a given volume. 8: AWS or Kubernetes or your favorite framework all come with some real cool capabilities and also some real trade-offs. What they're built to do makes some things much easier, and some things much harder. Do note that when you’re building something for the first time on a schedule, prioritizing to deliver a minimal first set of features also acts as an inherent optimization phase: what you choose to build and leave for later fits that same trade-off pattern. 9: This is similar to something called the Law of Fluency, which states that well-adapted cognitive work occurs with a facility that belies the difficulty of resolving demands and balancing dilemmas. While the law of fluency works at the individual cognitive level, I tend to assume it also shows up at larger organizational or system levels as well. 10: Rule- and Role-retreat may also be seen when people get overloaded, but won't deviate or adjust their plans to new circumstances. This "failure to adapt" can also contribute to incidents, and is one of the reasons why some forms of deviations have to be considered positive for the system. 11: Most of the information in this section came from Dr. Richard I. Cook, explaining the concept in a group discussion, a few years before his passing. 12: this isn’t purely a tooling decision; you also make this type of call every time you choose to refactor code to create an abstraction instead of copy/pasting bits of it around. 13: I believe but can't prove that there's also a tenuous but real path between the small-scale frictions, annoyances, and injustices we can let slip, and how they can be allowed to propagate and grow in greater systemic scales. There's always tremendously important work done at the local level, where people bridge the gap between what the system orders and what the world needs. If there are paths leading the feedback up from the local, they are critical to keeping things aligned. I'm unsure what the links between them are, but I like to think that small adjustments made by people with agency are part of a negative feedback loop partially keeping things in check.

This blog post originally appeared on the LFI blog but I decided to post it on my own as well. Every organization has to contend with limits: scarcity of resources, people, attention, or funding, friction from scaling, inertia from previous code bases, or a quickly shifting ecosystem. And of course there are more, like time, quality, effort, or how much can fit in anyone's mind. There are so many ways for things to go wrong; your ongoing success comes in no small part from the people within your system constantly navigating that space, making sacrifice decisions and trading off some things to buy runway elsewhere. From time to time, these come to a head in what we call a goal conflict, where two important attributes clash with each other. These are not avoidable, and in fact are just assumed to be so in many cases, such as "cheap, fast, and good; pick two". But somehow, when it comes to more specific details of our work, that clarity hides itself or gets obscured by the veil of normative judgments. It is easy after an incident to think of what people could have done differently, of signals they should have listened to, or of consequences they would have foreseen had they just been a little bit more careful. From this point of view, the idea of reinforcing desired behaviors through incentives, both positive (bonuses, public praise, promotions) and negative (demerits, re-certification, disciplinary reviews) can feel attractive. (Do note here that I am specifically talking of incentives around specific decision-making or performance, rather than broader ones such as wages, perks, overtime or hazard pay, or employment benefits, even though effects may sometimes overlap.) But this perspective itself is a trap. Hindsight bias—where we overestimate how predictable outcomes were after the fact—and its close relative outcome bias—where knowing the results after the fact tints how we judge the decision made—both serve as good reminders that we should ideally look at decisions as they were being made, with the information known and pressures present then.. This is generally made easier by assuming people were trying to do a good job and get good results; a judgment that seems to make no sense asks of us that we figure out how it seemed reasonable at the time. Events were likely challenging, resources were limited (including cognitive bandwidth), and context was probably uncertain. If you were looking for goal conflicts and difficult trade-offs, this is certainly a promising area in which they can be found. Taking people's desire for good outcomes for granted forces you to shift your perspective. It demands you move away from thinking that somehow more pressure toward succeeding would help. It makes you ask what aid could be given to navigate the situation better, how the context could be changed for the trade-offs to be negotiated differently next time around. It lets us move away from wondering how we can prevent mistakes and move toward how we could better support our participants. Hell, the idea of rewarding desired behavior feels enticing even in cases where your review process does not fall into the traps mentioned here, where you take a more just approach. But the core idea here is that you can't really expect different outcomes if the pressures and goals that gave them rise don't change either. During incidents, priorities in play already are things like "I've got to fix this to keep this business alive", stabilizing the system to prevent large cascades, or trying to prevent harm to users or customers. They come with stress, adrenalin, and sometimes a sense of panic or shock. These are likely to rank higher in the minds of people than “what’s my bonus gonna be?” or “am I losing a gift card or some plaque if I fail?” Adding incentives, whether positive or negative, does not clarify the situation. It does not address goal conflicts. It adds more variables to the equation, complexifies the situation, and likely makes it more challenging. Chances are that people will make the same decisions they would have made (and have been making continuously) in the past, obtaining the desired outcomes. Instead, they’ll change what they report later in subtle ways, by either tweaking or hiding information to protect themselves, or by gradually losing trust in the process you've put in place. These effects can be amplified when teams are given hard-to-meet abstract targets such as lowering incident counts, which can actively interfere with incident response by creating new decision points in people's mental flows. If responders have to discuss and classify the nature of an incident to fit an accounting system unrelated to solving it right now, their response is likely made slower, more challenging. This is not to say all attempts at structure and classification would hinder proper response, though. Clarifying the critical elements to salvage first, creating cues and language for patterns that will be encountered, and agreeing on strategies that support effective coordination across participants can all be really productive. It needs to be done with a deeper understanding of how your incident response actually works, and that sometimes means unpleasant feedback about how people perceive your priorities. I've been in reviews where people stated things like "we know that we get yelled at more for delivering features late than broken code so we just shipped broken code since we were out of time", or who admitted ignoring execs who made a habit of coming down from above to scold employees into fixing things they were pressured into doing anyway. These can be hurtful for an organization to consider, but they are nevertheless a real part of how people deal with exceptional situations. By trying to properly understand the challenges, by clarifying the goal conflicts that arise in systems and result in sometimes frustrating trade-offs, and by making learning from these experiences an objective of its own, we can hopefully make things a bit better. Grounding our interventions within a richer, more naturalistic understanding of incident response and all its challenges is a small—albeit a critical one—part of it all.

From time to time, people ask me what I use to power my blog, maybe because they like the minimalist form it has. I tell them it’s a bad idea and that I use the Erlang compiler infrastructure for it, and they agree to look elsewhere. After launching my notes section, I had to fully clean up my engine. I thought I could write about how it works because it’s fairly unique and interesting, even if you should probably not use it. The Requirements I first started my blog 14 years ago. It had roughly the same structure as it does at the time of writing this: a list of links and text with nothing else. It did poorly with mobile (which was still sort of new but I should really work to improve these days), but okay with screen readers. It’s gotta be minimal enough to load fast on old devices. There’s absolutely nothing dynamic on here. No JavaScript, no comments, no tracking, and I’m pretty sure I’ve disabled most logging and retention. I write into a void, either transcribing talks or putting down rants I’ve repeated 2-3 times to other people so it becomes faster to just link things in the future. I mostly don’t know what gets read or not, but over time I found this kept the experience better for me than chasing readers or views. Basically, a static site is the best technology for me, but from time to time it’s nice to be able to update the layout, add some features (like syntax highlighting or an RSS feed) so it needs to be better than flat HTML files. Internally it runs with erlydtl, an Erlang implementation of Django Templates, which I really liked a decade and a half ago. It supports template inheritance, which is really neat to minimize files I have to edit. All I have is a bunch of files containing my posts, a few of these templates, and a little bit of Rebar3 config tying them together. There are some features that erlydtl doesn’t support but that I wanted anyway, notably syntax highlighting (without JavaScript), markdown support, and including subsections of HTML files (a weird corner case to support RSS feeds without powering them with a database). The feature I want to discuss here is “only rebuild what you strictly need to,” which I covered by using the Rebar3 compiler. Rebar3’s Compiler Rebar3 is the Erlang community’s build tool, which Tristan and I launched over 10 years ago, a follower to the classic rebar 2.x script. A funny requirement for Rebar3 is that Erlang has multiple compilers: one for Erlang, but also one for MIB files (for SNMP), the Leex syntax analyzer generator, and the Yecc parser generator. It also is plugin-friendly in order to compile Elixir modules, and other BEAM languages, like LFE, or very early versions of Gleam. We needed to support at least four compilers out of the box, and to properly track everything such that we only rebuild what we must. This is done using a Directed Acyclic Graph (DAG) to track all files, including build artifacts. The Rebar3 compiler infrastructure works by breaking up the flow of compilation in a generic and specific subset. The specific subset will: Define which file types and paths must be considered by the compiler. Define which files are dependencies of other files. Be given a graph of all files and their artifacts with their last modified times (and metadata), and specify which of them need rebuilding. Compile individual files and provide metadata to track the artifacts. The generic subset will: Scan files and update their timestamps in a graph for the last modifications. Use the dependency information to complete the dependency graph. Propagate the timestamps of source files modifications transitively through the graph (assume you update header A, included by header B, applied by macro C, on file D; then B, C, and D are all marked as modified as recently as A in the DAG). Pass this updated graph to the specific part to get a list of files to build (usually by comparing which source files are newer than their artifacts, but also if build options changed). Schedule sequential or parallel compilation based on what the specific part specified. Update the DAG with the artifacts and build metadata, and persist the data to disk. In short, you build a compiler plugin that can name directories, file extensions, dependencies, and can compare timestamps and metadata. Then make sure this plugin can compile individual files, and the rest is handled for you. The blog engine Since I’m currently the most active Rebar3 maintainer, I’ve definitely got to maintain the compiler infrastructure described earlier. Since my blog needed to rebuild the fewest static files possible and I already used a template compiler, plugging it into Rebar3 became the solution demanding the least effort. It requires a few hundred lines of code to write the plugin and a bit of config looking like this: {blog3r,[{vars,[{url,[{base,"https://ferd.ca/"},{notes,"https://ferd.ca/notes/"},{img,"https://ferd.ca/static/img/"},...]},%% Main site{index,#{template=>"index.tpl",out=>"index.html",section=>main}},{index,#{template=>"rss.tpl",out=>"feed.rss",section=>main}},%% Notes section{index,#{template=>"index-notes.tpl",out=>"notes/index.html",section=>notes}},{index,#{template=>"rss-notes.tpl",out=>"notes/feed.rss",section=>notes}},%% All sections' pages.{sections,#{main=>{"posts/","./",[{"Mon, 02 Sep 2024 11:00:00 EDT","My Blog Engine is the Erlang Build Tool","blog-engine-erlang-build-tool.md.tpl"},{"Thu, 30 May 2024 15:00:00 EDT","The Review Is the Action Item","the-review-is-the-action-item.md.tpl"},{"Tue, 19 Mar 2024 11:00:00 EDT","A Commentary on Defining Observability","a-commentary-on-defining-observability.md.tpl"},{"Wed, 07 Feb 2024 19:00:00 EST","A Distributed Systems Reading List","distsys-reading-list.md.tpl"},...]},notes=>{"notes/","notes/",[{"Fri, 16 Aug 2024 10:30:00 EDT","Paper: Psychological Safety: The History, Renaissance, and Future of an Interpersonal Construct","papers/psychological-safety-interpersonal-construct.md.tpl"},{"Fri, 02 Aug 2024 09:30:00 EDT","Atomic Accidents and Uneven Blame","atomic-accidents-and-uneven-blame.md.tpl"},{"Sat, 27 Jul 2024 12:00:00 EDT","Paper: Moral Crumple Zones","papers/moral-crumple-zones.md.tpl"},{"Tue, 16 Jul 2024 19:00:00 EDT","Hutchins' Distributed Cognition in The Wild","hutchins-distributed-cognition-in-the-wild.md.tpl"},...]}}}]}. And blog post entry files like this: {% extends "base.tpl" %} {% block content %} <p>I like cats. I like food. <br /> I don't especially like catfood though.</p> {% markdown %} ### Have a subtitle And then _all sorts_ of content! - lists - other lists - [links]({{ url.base }}section/page)) - and whatever fits a demo > Have a quote to close this out {% endmarkdown %} {% endblock %} These call to a parent template (see base.tpl for the structure) to inject their content. The whole site gets generated that way. Even compiler error messages are lifted from the Rebar3 libraries (although I haven't wrapped everything perfectly yet), with the following coming up when I forgot to close an if tag before closing a for loop: $ rebar3 compile ===> Verifying dependencies... ===> Analyzing applications... ===> Compiling ferd_ca ===> template error: ┌─ /home/ferd/code/ferd-ca/templates/rss.tpl: │ 24 │ {% endfor %} │ ╰── syntax error before: "endfor" ===> Compiling templates/rss.tpl failed As you can see, I build my blog by calling rebar3 compile, the same command as I do for any Erlang project. I find it interesting that on one hand, this is pretty much the best design possible for me given that it represents almost no new code, no new tools, and no new costs. It’s quite optimal. On the other hand, it’s possibly the worst possible tool chain imaginable for a blog engine for almost anybody else.

2024/05/30 The Review Is the Action Item I like to consider running an incident review to be its own action item. Other follow-ups emerging from it are a plus, but the point is to learn from incidents, and the review gives room for that to happen. This is not surprising advice if you’ve read material from the LFI community and related disciplines. However, there are specific perspectives required to make this work, and some assumptions necessary for it, without which things can break down. How can it work? In a more traditional view, the system is believed to be stable, then disrupted into an incident. The system gets stabilized, and we must look for weaknesses that can be removed or barriers that could be added in order to prevent such disruption in the future. Other perspectives for systems include views where they are never truly stable. Things change constantly; uncertainty is normal. Under that lens, systems can’t be forced into stability by control or authority. They can be influenced and adapt on an ongoing basis, and possibly kept in balance through constant effort. Once you adopt a socio-technical perspective, the hard-to-model nature of humans becomes a desirable trait to cope with chaos. Rather than a messy variable to stamp out, you’ll want to give them more tools and ways to keep all the moving parts of the subsystems going. There, an incident review becomes an arena where misalignment in objectives can be repaired, where strategies and tactics can be discussed, where mental models can be corrected and enriched, where voices can be heard when they wouldn’t be, and where we are free to reflect on the messy reality that drove us here. This is valuable work, and establishing an environment where it takes place is a key action item on its own. People who want to keep things working will jump on this opportunity if they see any value in it. Rather than giving them tickets to work on, we’re giving them a safe context to surface and discuss useful information. They’ll carry that information with them in the future, and it may influence the decisions they make, here and elsewhere. If the stories that come out of reviews are good enough, they will be retold to others, and the organization will have learned something. That belief people will do better over time as they learn, to me, tends to be worth more than focusing on making room for a few tickets in the backlog. How can it break down? One of the unnamed assumptions with this whole approach is that teams should have the ability to influence their own roadmap and choose some of their own work. A staunchly top-down organization may leverage incident reviews as a way to let people change the established course with a high priority. That use of incident reviews can’t be denied in these contexts. We want to give people the information and the perspectives they need to come up with fixes that are effective. Good reviews with action items ought to make sense, particularly in these orgs where most of the work is normally driven by folks outside of the engineering teams. But if the maintainers do not have the opportunity to schedule work they think needs doing outside of the aftermath of an incident—work that is by definition reactive—then they have no real power to schedule preventive work on their own. And so that’s a place where learning being its own purpose breaks down: when the learnings can’t be applied. Maybe it feels like “good” reviews focused on learning apply to a surprisingly narrow set of teams then, because most teams don’t have that much control. The question here really boils down to “who is it that can apply things they learned, and when?” If the answer is “almost no one, and only when things explode,” that’s maybe a good lesson already. That’s maybe where you’d want to start remediating. Note that even this perspective is a bit reductionist, which is also another way in which learning reviews may break down. By narrowing knowledge’s utility only to when it gets applied in measurable scheduled work, we stop finding value outside of this context, and eventually stop giving space for it. It’s easy to forget that we don’t control what people learn. We don’t choose what the takeaways are. Everyone does it for themselves based on their own lived experience. More importantly, we can’t stop people from using the information they learned, whether at work or in their personal life. Lessons learned can be applied anywhere and any time, and they can become critically useful at unexpected times. Narrowing the scope of your reviews such that they only aim to prevent bad accidents indirectly hinders creating fertile grounds for good surprises as well. Going for better While the need for action items is almost always there, a key element of improving incident reviews is to not make corrections the focal point. Consider the incident review as a preliminary step, the data-gathering stage before writing down the ideas. You’re using recent events as a study of what’s surprising within the system, but also of how it is that things usually work well. Only once that perspective is established does it make sense to start thinking of ways of modifying things. Try it with only one or two reviews at first. Minor incidents are usually good, because following the methods outlined in docs like the Etsy Debriefing Facilitation Guide and the Howie guide tends to reveal many useful insights in incidents people would have otherwise overlooked as not very interesting. As you and your teams see value, expand to more and more incidents. It also helps to set the tone before and during the meetings. I’ve written a set of “ground rules” we use at Honeycomb and that my colleague Lex Neva has transcribed, commented, and published. See if something like that could adequately frame the session.. If abandoning the idea of action items seems irresponsible or impractical to you, keep them. But keep them with some distance; the common tip given by the LFI community is to schedule another meeting after the review to discuss them in isolation. iiii At some point, that follow-up meeting may become disjoint from the reviews. There’s not necessarily a reason why every incident needs a dedicated set of fixes (longer-term changes impacting them could already be in progress, for example), nor is there a reason to wait for an incident to fix things and improve them. That’s when you decouple understanding from fixing, and the incident review becomes its own sufficient action item.