A builder’s playbook for products designed for both humans and AI agents
slobodan_ · reddit · 2026-07-23
A builder shares lessons from shipping a product designed for both humans and AI agents from day one.
- Everything a person can do in the browser, an agent can do via API or an MCP server.
- The key takeaway: do not use an LLM to check another LLM; use deterministic checks instead.
- The team uses lint plugins, Claude Code hooks, CI word bans, descriptive error messages, monorepo-aware tooling, custom agent tools, and background jobs to make agent workflows reliable.
- The post argues that building with AI still takes weeks or months, but the process is now far more tractable.
More from coding & agent
- LLMs may help most in the manual refinement phase of decompilation — OwariDa · 2026-07-23
- Open-source course says the harness, not the model, can push a coding agent into the top 5 — strickvl · 2026-07-23
- TERSE proposes an open semantic state protocol for AI apps and agents — Defiant-Juice-2745 · 2026-07-23
- A Codex Remote meme about debugging an agent while it’s still running — Dimillian · 2026-07-23
- Claude Code orchestration layer routes subtasks to cheap workers and escalates on verified failure — StolenDrinks · 2026-07-23
- Reddit user says Codex is silently downgrading from GPT 5.6 Sol-High to Luna-Low — SweetGirlKatie · 2026-07-23