How to Roll Out Comprehend-First AI Coding to Your Team
For Engineering leads driving AI adoption · Based on Priscila Andre's Comprehend-First AI Coding Skill
// TL;DR
Worried your team is shipping slop code from unchecked AI generation? The Comprehend-First skill gives you a standard to enforce: comprehend first, code later. Roll out the Catch Me Up prompt as a shared Markdown asset encoding six exploration modes, add a mandatory comprehension gate between research and planning, and coach engineers to audit their own AI usage. Most assume they're mostly generating; audits often show ~67% comprehension. Standardizing this turns your team into disciplined Agent Managers who ship defensible keynote code instead of unaccountable slop.
Why is my team shipping slop code with AI?
Because the default AI workflow skips comprehension and jumps straight to generation. Slop code — shipped into a production codebase without genuine understanding — is the primary source of bugs in AI-assisted teams. When engineers say 'the AI wrote it' as an excuse, you've lost the accountability that keeps a codebase healthy. The Comprehend-First skill gives you a concrete standard to enforce: never ship code into a codebase you don't understand.
The reframing is powerful. Most engineers assume their AI usage is dominated by code generation. But when the creator audited real prompt history, ~67% was comprehension and only ~2% was generation. Your team is probably already doing comprehension work — they just aren't doing it deliberately or well. Making it explicit is where the leverage is.
How do I standardize this across the team?
Turn Catch Me Up into a shared asset. Store it as a local Markdown file that encodes the six exploration modes — Architecture, Convention, Feature, Trace, Syntax, Testing, History — and the prompt structure: [Role statement] + [Catch me up on X] + [specific clarifying question]. When every engineer invokes comprehension the same way, you get consistency, reviewability, and a common vocabulary.
Then codify the workflow: Research → Comprehend → Plan → Implement. The critical addition to the popular three-phase model is the mandatory Comprehension gate between Research and Planning. Make it a norm that no one plans or implements against research they can't personally explain. This single gate prevents the most expensive AI failures — wrong-direction agent runs and slop merges.
How do I get engineers to actually adopt it?
Have each engineer audit their own AI usage. Reviewing real prompt history and categorizing prompts by intent surfaces the gap between assumed and actual patterns. Once someone sees they're mostly comprehending, they invest in better comprehension prompts and tooling instead of optimizing generation. Ownership beats mandate.
Reinforce the culture principles: there are no dumb questions for AI — it's the cheapest senior engineer available, so juniors should ask it everything they'd hesitate to ask a busy colleague. And frame the aspiration positively: the goal is keynote code, work you'd be proud to present publicly, not slop you can't defend.
How does this change the role of my senior engineers?
It formalizes the Agent Manager role. Senior work shifts toward comprehension, direction-setting, and quality control rather than raw implementation. Your best engineers become the ones who build accurate mental models fastest, verify agent output, and steer agents away from drift. Reward that behavior in reviews and promotions — it's the durable skill as agentic coding scales.
Measure adoption by asking whether every merged change can be explained in plain language by its author, and whether reviewers used comprehension modes before approving unfamiliar code. Those two checks operationalize the whole skill.
Next step
Create a shared Catch Me Up Markdown file this week, add a 'comprehend before you plan' step to your team's definition of done, and have each engineer audit their AI prompt history. Set one rule: no code merges into a real codebase unless its author can explain what it does and why.
// FREQUENTLY ASKED QUESTIONS
How do I measure whether Comprehend-First is working on my team?
Use two operational checks: can every merged change be explained in plain language by its author, and did reviewers use comprehension modes before approving unfamiliar code? Also track AI usage audits — a healthy shift is engineers deliberately investing in comprehension prompts rather than assuming generation dominates. Fewer wrong-direction agent runs and slop merges are the downstream signal.
What's the single most important change to enforce?
Add the mandatory Comprehension gate between Research and Planning. The popular three-phase model (Research, Planning, Implementation) lets agents plan against research no one personally understood. Requiring that engineers can explain research findings in plain language before planning prevents the most expensive AI failures — wrong-direction runs and slop code merges.
How do I get skeptical senior engineers to buy in?
Have them audit their own prompt history. Most assume they're mostly generating code, but audits often reveal ~67% comprehension. Seeing their real pattern reframes AI as a comprehension tool and a lever for the emerging Agent Manager role — senior work centered on understanding, steering, and quality control rather than raw implementation.