# No Coding Before 10am A startup I work with just threw out their entire playbook. They've been moving fast, but over the past month, their way of working completely collapsed because of Claude Code and Codex. So they held a war room meeting and rebuilt how to operate from scratch. Their first new rule: **No coding before 10am.** For the past two decades, engineering culture has been focused on maximizing time spent writing code. Cut meetings. Lock your calendar. Block anything that takes engineers away from their keyboards. This team is doing the complete opposite. Every morning, engineers now pair-prompt together: they sit together, draft prompts, define goals, and set up the Agent so it can succeed. Only after that does the Agent start working. Their playbook isn't "write code faster with AI." It's a total inversion. **The Agent, not the engineer, is now the doer. The engineer makes sure the Agent can do the work well.** Over the past decade, I've seen dozens of teams operate—from DoorDash to my own startups. What they've distilled is the clearest version I've seen of "how engineering actually works now." I asked if I could share. Here’s their playbook. --- **Agent is the Primary User** 1. **Design for the Agent, not for humans.** Every system, data store, naming convention, and knowledge artifact should be designed with the AI Agent as the primary consumer. Humans interact with systems through the Agent as much as possible. 2. **Code is context, not a library.** Agents read code to understand what it does, then generate their own version. Don't optimize for "cross-human code reuse." Optimize for "Agent comprehensibility of code." Code itself is documentation. 3. **Data is the real interface.** The correct interface between two components is a well-structured data artifact, not a function call. Clean data lets Agents compose systems on their own without being told how. 4. **Maximize Agent utilization.** If the team is commuting and nothing is running, that's waste. Agents should work at night, during commutes, during meetings, asynchronously. The most expensive thing in the system is the Agent/compute sitting idle waiting for a human. --- **How to Spec and Build** 1. **Goals and constraints first, above all else.** Before building anything: write the goal in one sentence, list constraints, define success criteria. If you can't state the goal in one sentence, you don't understand the problem enough to build it. 2. **Spec the outcome, not the process.** AI will figure out how. You judge the output against the objective function. This replaces traditional PRDs. Write objective functions, not implementation plans. 3. **Define rules, not structures.** Don't over-specify schemas and formats. Set naming conventions, metadata requirements, and versioning rules. Let the Agent figure out the rest. 4. **Review the output, not the code.** Don't read every line of code the Agent wrote. Test the code against the goal. Pass? Ship it. Fail? Redefine goals and constraints. Code review, as we once knew it, is overhead this system no longer needs. 5. **When you build a new way, kill the old way.** Don't run parallel implementations. Remove old code paths immediately. The codebase is the Agent's context. Every dead path is noise that reduces Agent performance. 6. **Think systemically.** If you do something manually more than twice, automate it. If a human is still repeating a task, the system isn't set up right. The goal: set things up, let them run, check the output, move on. --- **Working Together** 1. **No coding before 10am.** Hands off the keyboard. The first hour or two each morning is for talking, aligning, and drafting prompts together. Once the team agrees on what to build and how to set up the Agent, then you start coding—let the Agent work. 2. **Optimize for time, not tokens.** If spending 10x the tokens saves a day, spend them. The bottleneck is human decision time, not compute cost. 3. **Individual autonomy, shared interfaces.** Everyone uses their own IDE, their own prompting style, their own workflow. What's standardized: data schemas, goal specifications, component responsibilities. Everything else is personal choice. 4. **Call out anti-patterns immediately.** When you catch yourself or others slipping back into old habits—designing for humans instead of Agents, accumulating dead code, skipping specs—call it out right away. Old habits accumulate fast. 5. **Assume everything changes in three months.** Technology is changing every month. Every decision you make today will soon be wrong. Build modularly. Minimize lock-in at every layer. --- In six months, there will be two kinds of engineering teams: **those who rebuilt their way of working from first principles**, and **those still trying to shove Agents into the old playbook**. The latter will be left far behind by teams half their size. If you run an engineering team and haven't had your own version of this war room meeting yet, go have it. Throw out the old playbook. Write a new one. --- Original: Michael Bloch, *No Coding Before 10am*, Feb 2026 https://michaelxbloch.substack.com/p/no-coding-before-10am… #AIEngineering #Agent #SoftwareEngineering #WorkStyle