Choose a useful job for AI
Practical implementation patterns and draft guides by Naturate. Start with a problem, then inspect the example and its limitations.
Use the library to connect an existing service, coordinate a small workflow, build an admin tool, or improve an AI feature you already run.
Each guide should explain the result, the implementation and the failure cases. In production means evidence from its original implementation, not independent validation. Draft means the guide still needs implementation or host testing. Read the evidence section before treating a pattern as proven.
The catalog below is generated from the recipe pages themselves. For the format and evidence standards, see How to use these recipes.
Connect AI to your product
Let agents use what you already have, through a boundary you control.
Make an existing service usable by AI agents
Expose a small, useful part of your service through MCP. Start with two read-only tools, not a copy of your entire backend.
Build admin tools that humans and agents can share
Give people a clear review surface and agents a small set of typed actions. Keep permissions and business rules in one backend.
Ground your agents in the world outside
Add Pointmoon to an agent that needs to know what is happening at a real place right now. One URL, no key. Every fact arrives with its source and time, or the agent is told plainly that nothing is known.
Make AI output reliable
For features where a wrong or late answer costs you trust.
Keep an AI feature reliable in production while you keep changing the prompt
You tweak the prompt and quietly break something else; the model makes things up; the provider hiccups and your feature goes dark. Here's how to run an AI feature so none of that happens, change prompts without a deploy, auto-check every output, never hard-fail, and prove a change is better before it ships.
Prove an AI change is better before it ships
You changed a prompt or a model and it looks better on the three examples you tried. Here is how to find out whether it is better on the inputs your users actually send, and what it broke, before anyone sees it.
Make an LLM talk about the real world without making things up
Your LLM describes something real, the weather, a location, live data, and lies with total confidence. Here's the discipline where the model only ever phrases facts a data layer actually asserts: every fact carries provenance, unsourced and stale claims are stripped, and when there's nothing solid it says nothing instead of inventing.
Make an AI voice feature start playing in seconds, not after a long wait
You hand LLM-generated narration to a text-to-speech provider and users stare at a loading spinner for tens of seconds before anything plays, or you reach for a faster voice model and it drops your pauses, mispronounces markup, or wanders accent mid-clip. Here's the pipeline that streams audio starting in seconds, keeps pauses and loudness consistent across voice tiers, and degrades gracefully instead of hanging.
Teach an assistant when to say nothing
An assistant that speaks up whenever it can is muted within a week. Here is how to build one whose default is silence, that speaks only when the moment has earned it, and that says one useful thing when it does.
Run agents on real work
Several agents, real tasks, and a person who still decides what goes out.
Get Claude and ChatGPT working on the same task
Let one assistant produce an artifact and another review it. Share the result and its evidence, not an endless conversation.
Build an autonomous daily desk, not an autonomous company
Turn a small queue into reviewable drafts on a schedule. Bound the work, keep consequential decisions with a person, and make missed runs visible.
Give your AI agents a shared memory that survives every session
Your agents forget everything when a session ends, re-derive settled facts, and contradict decisions nobody wrote down. Here's how to give them a durable, file-based shared brain, one place agents read at session start and write back at session end, so knowledge compounds instead of resetting.
Run a fleet of AI agents on real work without babysitting them
Several agents at once, on a schedule, while you're not watching, and it keeps going wrong: they clobber each other, run on your laptop with your credentials, stop to ask permission for everything, or fail silently for days. Here's the pattern that keeps a fleet safe and honest: isolated ephemeral workers, output-gated not permission-gated, fan-out caps, and a dumb watchdog that pages on silence.
Give each agent a lane
Ask a general agent to run your marketing and it does a bit of everything and owns nothing. Here is how to split the work into agents that each own one job and one number, and know what is not theirs.
Run two coding agents in one repository without losing work
Two agents, or an agent and a person, working the same repository will overwrite each other unless each has its own working tree. Here is the setup that keeps everyone's work, and how to get work back when it is lost anyway.
Let agents ship without breaking production
When agents merge often, the code that was tested and the code that gets deployed drift apart. Tie every check to one commit and its build, deploy exactly that, and confirm what is live.
Stop agents overstating what they did
An agent says "done" and it is on a branch nobody can see. It says "verified" and it read the code. Here is how to make an agent's report mean what it says, and how to catch the ones that do not.