> ## Content Index
> Fetch the complete content index at: https://www.antoinebuteau.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# When Heroics Become the Operating System
- URL: https://www.antoinebuteau.com/when-heroics-become-the-operating-system/
- Published: 2026-10-02T04:10:22.000Z
- Updated: 2026-10-02T04:10:22.000Z
- Description: Hero culture is what happens when a growing company borrows scale from a few people’s private context—and rewards them for repeatedly rescuing the system.
- Author: Antoine Buteau
- Tags: Leadership, Management, Organizational Design

Hero culture rarely begins as a cultural choice.

It begins because the company is growing, everything is moving too quickly, and a handful of people know how the pieces fit together. They remember why a customer promise was made, which shortcut is dangerous, where the system is fragile, and who can unblock a decision at 11 p.m.

When something catches fire, they fix it. Then they fix the next thing. Their value is obvious, the company is grateful, and the story becomes self-reinforcing: these are the people who get things done.

For a while, this works.

Then the company discovers that it did not build an operating system. It borrowed scale from a few people’s nervous systems.

At scale, hero culture becomes a context-distribution problem, an incentive problem, and eventually a capacity problem. Growing out of it does not require becoming slow or hostile to exceptional effort. It requires converting private judgment into organizational capability before the people carrying the company become the constraint on it.

## The hero is usually a symptom

You can usually spot the pattern from the meeting list. The same people appear in every difficult situation. They join the most important customer call, debug the incident, rewrite the plan, resolve the cross-functional dispute, and explain the historical reason a seemingly irrational process exists.

Everyone else learns to route ambiguity toward them.

This looks like excellence because the rescue is real. The mistake is assuming the rescue proves the system is healthy.

James Rosen-Birch describes a more useful causal loop in his analysis of [behaviours that kill companies](https://twitter.com/provisionalidea/status/1524038465582735360?ref=antoinebuteau.com). Weak communication and repeated work lead to lost institutional knowledge. Missing infrastructure creates operations debt. That debt produces fires. Firefighting shortens the company’s time horizon, leaving even less time to build the infrastructure that would prevent the next fire.

The hero stands in the middle of that loop. They are not necessarily causing it. They are making it survivable.

Blaming the individual punishes someone for compensating for structural weakness. Treating their rescues as the solution leaves that weakness untouched.

Instead of asking people to stop being heroes, ask what keeps requiring rescue and why the same person has to perform it.

## Growth makes concentrated context look like talent

Early companies run on proximity. Founders and early employees share conversations, customers, mistakes, and decisions. They can coordinate with fragments: a Slack message, a look across the room, or a five-minute call.

As the company grows, knowledge expands while each person sees a smaller share of it. Conor Brennan-Burke describes the point when [the company stops fitting inside anyone’s head](https://twitter.com/contextconor/status/2085022975414219258/?rw%5Ftt%5Fthread=True&ref=antoinebuteau.com). More people create more knowledge, but they also create more boundaries that knowledge must cross.

Veterans have an advantage because they carry the missing links. They remember what the company decided and why. They understand which rules are principles, which are temporary workarounds, and which can be broken safely.

That makes them faster than everyone else. But speed based on inaccessible context is not yet a scalable capability.

[Railway described this failure mode](https://blog.railway.com/p/how-we-work-volume-iv?ref=antoinebuteau.com) when experienced employees worked alone on increasingly difficult systems. The work was impressive, but the company had “at best, half a person” of backup context. New hires could not find meaningful on-ramps. When they attempted the hardest work without enough context, mistakes reduced the veterans’ trust and pushed ownership back toward the same small group.

Soon, the work begins to route itself:

1. The veteran has the most context, so they take the hardest work.
2. Others receive safer or less consequential work, so they gain context slowly.
3. The gap between the veteran and the team widens.
4. Delegation feels increasingly risky.
5. The next critical problem returns to the veteran.

The organization then misreads accumulated access to context as an irreplaceable individual trait.

## The hidden bargain eventually breaks

Hero culture creates a bargain between the company and its most relied-upon people.

The company gets speed, continuity, and emergency judgment without first building durable systems. The hero gets trust, influence, recognition, and the satisfaction of being essential.

Both sides can become attached to the arrangement.

Leaders delay hard operating work because a strong person keeps catching the failures. Heroes may resist handing off work because delegation initially feels slower, produces imperfect results, and reduces the identity reward of being the person who always saves the day. Teams learn that the fastest route through uncertainty is escalation rather than judgment.

The bargain breaks when demand grows faster than the hero’s capacity.

Leo Polovets’s essay on [becoming your future self](https://codingvc.com/becoming-your-future-self/?ref=antoinebuteau.com) describes how strong early employees become bottlenecks when they remain attached to the work that made them successful. A CTO who still wants to code critical features while managing a growing organization does neither job well enough. Teammates wait for work the CTO no longer has time to complete.

At that point, hero culture stops increasing velocity. It creates a queue.

The warning signs are practical:

- important work waits for one person to review or approve it;
- incidents are resolved without a durable change to the system;
- the same issues recur under different names;
- new hires spend months on low-risk work because meaningful ownership feels unsafe;
- leaders say they want delegation but keep reversing delegated decisions;
- vacations create anxiety because context has no backup owner;
- the people with the most context have the least time to transfer it;
- prevention work is repeatedly deferred because current delivery is “too urgent.”

When several of these are true, the company does not have a hero problem. It has an operating model that consumes heroes.

## Replace dependency without adding drag

Companies often respond too late and too bluntly. After enough misses, they add approval gates, committees, templates, and mandatory processes. Informal judgment is replaced with formal control.

That may reduce certain errors, but it also makes the organization slower and strips authority from the people closest to the work.

[Netflix’s culture model](https://jobs.netflix.com/culture?ref=antoinebuteau.com) frames the alternative as “context, not control.” Leaders create the information and judgment conditions for independent decisions. They stay close enough to learn where context is weak without retaking every decision. Better context lets more people make sound calls on their own.

Good structure distributes sound judgment with as little control overhead as possible.

In practice, this looks like:

- clear ownership rather than collective concern;
- decision context rather than approval chains;
- escalation thresholds rather than constant permission-seeking;
- written history rather than oral tradition;
- rehearsed backups rather than nominal backups;
- recurring prevention capacity rather than occasional cleanup days.

Structure should remove avoidable ambiguity. It should not remove initiative.

## How to grow out of hero culture

Start by treating every rescue as operational evidence.

### 1\. Build a rescue ledger

For 30 days, record consequential interventions. What happened? Who was pulled in? What context did they possess that others did not? Was the issue resolved, or was the immediate symptom contained?

Do not turn this into a blame exercise. Look for repeated dependencies. If one person appears in six unrelated incidents, the common factor may not be that they are uniquely brilliant. Decision rights, system knowledge, customer history, or escalation paths may all converge on them.

### 2\. Separate incident command from permanent ownership

During a real fire, speed matters. Put the most capable person in charge if necessary. But assign a second owner to the aftermath: document the decision path, identify the missing control or context, and ensure prevention work enters a real backlog with capacity.

Otherwise, incident response becomes performance. The company applauds the save and forgets to change the conditions that required it.

### 3\. Capture the reasoning behind the procedure

Documentation often fails because it records steps without judgment. A checklist says what to do in the normal case. It rarely explains which signal changes the decision, what tradeoff matters, or when the rule should be broken.

Dan Hockenmaier’s account of [Faire’s operating system](https://twitter.com/danhockenmaier/status/2089777541506896007/?rw%5Ftt%5Fthread=True&ref=antoinebuteau.com) describes a writing culture where major problems and decisions are captured in documents before meetings. The value is not administrative neatness. The company creates a reusable record of the problem, comments, response, and decision. More people can learn from the same insight, and old questions do not need to be reconstructed from memory.

For every important decision, capture four things: the situation, the governing constraints, the alternatives considered, and the reason for the choice. That is the material another person needs to exercise judgment later.

### 4\. Delegate in stages

Dumping a critical responsibility on someone without context is abandonment, not delegation. Keeping it forever is dependency.

A better sequence is:

1. **Observe:** the learner watches the decision and hears the reasoning.
2. **Recommend:** the learner develops a recommendation; the current owner decides.
3. **Decide with review:** the learner decides before execution, with a bounded check.
4. **Decide and inform:** the learner acts, then communicates the result.
5. **Own:** the learner becomes the primary owner and develops the next backup.

Steven Sinofsky argues that as functions multiply, [systematic delegation and decision-making become mandatory](https://medium.learningbyshipping.com/a-leader-s-guide-to-deciding-what-when-and-how-to-decide-f6caf659fa7e?ref=antoinebuteau.com). Assigning authority is easy. Making the context and consequences of the decision legible is harder.

### 5\. Create real on-ramps to consequential work

Teams cannot build redundancy if every meaningful problem remains too important for anyone new to touch.

Railway’s response was deliberately simple: stop polishing every task and preserve meaningful “good first” work that let others gain context. Outside engineering, the next owner might run part of the customer review, draft the operating decision, lead the incident retrospective, or make a bounded call where the downside is reversible.

Redundancy is built through reps, not org charts.

### 6\. Change what earns status

If promotion, praise, and access flow to the person who fixes visible crises, the company will produce more crisis fixers. Prevention is quieter. Good handoffs, simpler systems, clearer interfaces, and better-trained successors are easy to overlook precisely because they make drama disappear.

Review leaders on whether their area can operate without their constant intervention. Celebrate the person who eliminated an escalation class, created a capable backup, or made a complex decision reusable.

Strong people remain important. Their contribution now compounds through others.

### 7\. Protect capacity for prevention

Firefighting consumes the time required to stop firefighting. Breaking that loop requires an explicit allocation of capacity, not a hope that infrastructure work will happen after urgent work is finished.

Kellan Elliott-McCrea’s [organizational-design diagnostic](https://kellanem.com/notes/whys-of-organizational-design?ref=antoinebuteau.com) makes the tradeoff clear: when senior engineers spend all their time firefighting, the system often lacks enough slack for training and tool-building. Without protected capacity, the organization keeps buying short-term delivery with long-term fragility.

Prevention work needs an owner, a budget, and inspection. If it is always optional, it will always lose.

## The test of a scalable hero

Exceptional people still matter. There will always be moments when someone sees the whole field, moves first, and carries more than their formal share. Growing out of hero culture does not mean suppressing that behavior.

It means changing its output.

The scalable hero solves the hard problem and leaves behind a stronger system, a clearer decision record, and another person who can handle the next version. Their value shows up in how well the company operates after they leave the room.

Rodolphe Barrere describes this as trying to [“fire himself” from the operating chain](https://betakit.com/why-potloc-founder-rodolphe-barrere-tries-to-fire-himself-every-six-months/?ref=antoinebuteau.com). Leaders whose companies are outgrowing informal coordination should aim for the same outcome: build the foundation, transfer the context, and remove themselves as dependencies so they can move to the next frontier.

The company has grown out of hero culture when the next crisis does not need the same rescuer—and when every rescue leaves fewer reasons to need one.

## Sources

- [What patterns of behaviour kill companies?](https://twitter.com/provisionalidea/status/1524038465582735360?ref=antoinebuteau.com), James Rosen-Birch
- [Intelligence Is Cheap. Coordination Is Not.](https://twitter.com/contextconor/status/2085022975414219258/?rw%5Ftt%5Fthread=True&ref=antoinebuteau.com), Conor Brennan-Burke
- [How We Work, Volume IV](https://blog.railway.com/p/how-we-work-volume-iv?ref=antoinebuteau.com), Railway
- [Becoming Your Future Self](https://codingvc.com/becoming-your-future-self/?ref=antoinebuteau.com), Leo Polovets
- [Netflix Culture](https://jobs.netflix.com/culture?ref=antoinebuteau.com), Netflix
- [How to Build a High-Growth Operating System](https://twitter.com/danhockenmaier/status/2089777541506896007/?rw%5Ftt%5Fthread=True&ref=antoinebuteau.com), Dan Hockenmaier
- [A Leader’s Guide to Deciding](https://medium.learningbyshipping.com/a-leader-s-guide-to-deciding-what-when-and-how-to-decide-f6caf659fa7e?ref=antoinebuteau.com), Steven Sinofsky
- [The 5 Whys of Organizational Design](https://kellanem.com/notes/whys-of-organizational-design?ref=antoinebuteau.com), Kellan Elliott-McCrea
- [Why Potloc’s Founder Tries to Fire Himself Every Six Months](https://betakit.com/why-potloc-founder-rodolphe-barrere-tries-to-fire-himself-every-six-months/?ref=antoinebuteau.com), Stefan Palios