How Anthropic Runs Crisis Projects: A Playbook for Technical Leaders

Some weeks at a company feel normal. Others feel like controlled chaos where everything clicks into place at impossible speed. Ben Kuhn, who’s been at Anthropic through several such weeks, just shared a detailed playbook for crisis project management—and it’s a masterclass in high-stakes technical leadership.

The Core Insight

The most productive weeks at Anthropic have been “crisis project management”—coordinating major, time-sensitive implementation or debugging efforts. But here’s the counterintuitive part: what made these weeks successful felt relatively straightforward. The secret wasn’t brilliance; it was relentless execution of fundamentals.

The real insight? Project management is extremely high-leverage in complex technical organizations, but it’s rarer than it needs to be. Most teams miss obvious opportunities because they’ve never seen a good playbook.

Why This Matters

Information Processing Is the Bottleneck

A key observation that changes everything: getting complete information is usually the hard part, often consuming a substantial fraction of the critical-path timeline.

Take this debugging timeline from a recent training run:
1. Chips delivered → tests run → unexpected poor performance
2. Escalate to compute partner → they staff debugging effort
3. Realize you gave them outdated benchmark → wrong target
4. Switch benchmark → share code → partner rolls out improvements
5. Test → still poor → repeat until good

Almost every step is information processing, not coding. Even the debugging itself was constrained by how quickly info could round-trip between teams.

The OODA Loop Is Everything

OODA: Observe, Orient, Decide, Act. For complex projects, this loop is the bottleneck. Ben’s tactics for running it faster:

Spend time on it: 6+ hours a day during crisis mode just on coordination. Information processing isn’t a “free action”—it’s the core job.

Communicate uncomfortably much: Multiple daily calls with counterparts. 9am and 6pm syncs during critical debugging. Constant bouncing between groups asking for updates.

Track biggest open questions: A living doc with ranked uncertainties. Resolving these is the priority list. Ideally, enough people to work on multiple in parallel.

Reorient frequently: Review priorities multiple times daily. Check that work is attacking the right things. Direction matters more than velocity.

The Plan for Victory

A detailed, concrete plan ending with the goal being achieved. Not because plans survive contact with reality, but because deviation from plan is how you know when to freak out.

The most common megaproject failure mode: not freaking out soon enough.

“We also massively underestimated a few components of the project, and because of this, we still ended up very crunched at the end.”

Even imperfect plans help. The failure wasn’t having a plan—it was underestimating tasks and not checking in frequently enough against it.

Key Takeaways

• Focus completely: Clear your schedule. 6+ hours/day isn’t too much for a crisis project. The top idea in your mind should be the project.

• Plans detect failure: The point of a detailed plan isn’t to follow it perfectly—it’s to know when things are going wrong.

• Overcommunicate: Repeat things way more often than feels natural. Everyone needs ambient awareness to make autonomous decisions.

• Break off subprojects: At ~10 people, delegate project management, not just execution. Best PMs are organized and goal-focused, not necessarily the strongest ICs.

• Keep goals simple: If a goal can fit in a Slack message while crisply describing the path to success, people can prioritize autonomously.

• Technical context matters: A common anti-pattern is having trusted decisionmakers delegate execution to lower-context PMs. Usually a false economy for critical projects.

Looking Ahead

What’s striking about this playbook is how unsexy it is. No revolutionary methodologies. No new frameworks. Just fundamentals executed with intensity and discipline.

The crisis project lessons translate directly to non-crisis work—just with less intensity. Carve out daily time for status checks, priority reviews, and broadcasts. Make the project a top priority even when nothing’s on fire.

The teams that excel at this aren’t more innovative. They’re more relentless about the basics that everyone knows but few execute consistently.

Tags: Project Management, Technical Leadership, OODA Loop, Team Coordination, Anthropic, Engineering Management

Based on analysis of How I’ve run major projects by Ben Kuhn