A Daily Cadence for Solo Dev + AI: Plan in the Morning, Check at Noon, Review at Night

The bottleneck moved: from typing speed to decision speed

After pairing with AI for a while, you'll hit a counterintuitive truth: writing code is no longer the bottleneck — your decision speed is. The AI can produce three days of output in a morning, but "is this approach right," "which first," "when is this bug fixed enough" — those decisions are yours alone. Without rhythm, the AI's output drowns you: five tasks opened in the morning, three wrong directions discovered in the afternoon, an evening spent cleaning up after the AI.

The fix isn't working harder; it's a fixed daily cadence. Weld decision points into the schedule so the AI's output gets accepted at fixed checkpoints instead of consuming your attention all day. I've used this rhythm for a long time — steal it wholesale.

Morning, 15 minutes: 1 goal + 3 verifiable tasks

Before touching code, spend 15 minutes writing today's battle plan. The format is rigid: one one-sentence goal ("today, users can complete the full register-to-checkout flow") + no more than 3 verifiable tasks (each must be describable in one sentence as "how I'll know it's done").

The key rule: task descriptions state acceptance criteria, never implementation plans. "Cache sessions in Redis" is a plan; "users stay logged in after page refresh" is acceptance. Let the AI decide plans — that's what it's best at; you own the criteria — that's non-transferable. Paste the plan to the AI, have it restate today's plan, and start only once aligned.

Noon checkpoint: judge acceptance criteria, not code

Take a mandatory 20-minute checkpoint at noon with one rule: check off acceptance criteria one by one; don't read code line by line. Standardize the AI's report format too: one sentence per task (done / not done / blocked where) + acceptance evidence (test screenshots, test output, demo links).

This habit kills the solo developer's biggest invisible killer: the slow boil of "good enough." Without checkpoints, morning holes explode in the afternoon; with them, a wrong direction costs at most half a day. Remember: you're the QA inspector, not the code reviewer — your time goes to judging "is it right," not "how was it written."

Evening, 10 minutes: review + write the "context handoff note"

Two things before shutdown. First, review three questions: which decision today was right (repeat it), which was wrong (where exactly), where did the AI waste the most tokens (how to avoid it tomorrow). Write it down, max five lines — this is your personal "AI collaboration playbook," worth more than any tutorial.

Second, write a "context handoff note" for tomorrow's you (and the AI): where things stand, which branch, which tests still fail, what to do first tomorrow. The AI's context window will clear and your memory will blur; this note is the key to being battle-ready in five minutes the next morning. Keep it at the project root as TODAY.md — tomorrow's first message is simply "read TODAY.md and continue."

Friday afternoon: pay down debt, open no new fronts

Reserve half a day weekly (I use Friday afternoon) for one thing: debt repayment, no new features. Have the AI: delete dead code and commented blocks from the week, fill in missing tests, update stale README sections, archive TODAY.md entries into a weekly note. Technical debt not cleaned on Friday collects triple interest on Monday.

One uncomfortable truth to close: many ask "which AI coding tool is best" but never "is my working rhythm right." The gap between tools is 10%; the gap between "having rhythm" and "not having rhythm" is 10x. AI amplifies your output — and your chaos. Rhythm first, tools second. Take this morning-noon-evening-plus-Friday frame and run with it.