February 15, 2026
Prompt and Circumstance
No Coding Before 10am
No Coding Before 10am: Bots take the wheel, devs sip coffee and rage
TLDR: A startup bans early-morning coding so engineers can brief AI bots that write the code later, flipping “humans build” into “humans direct.” Commenters split: some call it wasteful or awful, others crack jokes about government IT, and many worry the coder role is becoming a bot-wrangler — a sign of work changing fast.
A startup just told engineers to stop coding before 10am so they can spend mornings writing instructions for AI “agents” (bot workers) that do the typing later. In the post by Michael Bloch, the company flips the script: build systems for bots, judge results not line-by-line code, and delete old paths fast so the bots don’t get confused.
Cue the comments cage match. One camp calls it meetings with extra steps, rolling their eyes at “agents first” and the idea that spending more on AI compute saves time. As one critic asks, why burn cash “optimizing what isn’t the bottleneck”? Another dev says they do their best coding early and “get progressively dumber” as the day goes on — so wasting prime brain hours on prompts feels like a crime. Others simply call the whole thing “awful.”
Then the comedy crew shows up. Someone jokes you can get the same “no coding” vibe by joining a government IT shop. Another laments they can’t even get coworkers to use basic tools, suggesting many teams aren’t ready for bot-first anything. Still, fans of the idea say it’s the logical next step: let bots grind while humans set the goalposts. The culture clash is real — craft vs. control panel, coffee vs. prompts, and a lurking fear that the “coder” job is turning into “bot wrangler.”
Key Points
- •A startup overhauled its engineering workflow due to the impact of AI coding tools like Claude Code and Codex.
- •The team instituted a “no coding before 10am” rule to prioritize morning prompt sessions that define objectives and constraints for agents.
- •The playbook shifts to agents as primary executors of work, with systems designed for agent consumption and data as the main interface.
- •Specifications focus on outcomes via objective functions, with rules (naming, metadata, versioning) defined instead of rigid schemas.
- •Code review centers on testing outputs; old code paths are removed immediately; repeated manual tasks are automated.