Labs
Labs is the safe place to try things. Changes you make here do not touch your real work until you decide they should.
Think of it as a workshop next to your office. You can take something apart in the workshop without stopping the work happening next door.
What Labs is for #
- Testing a change before it affects your daily work
- Comparing two ways of doing the same job
- Building a reusable set of steps for something you do often
- Keeping useful abilities in one place
What is inside #
| Tab | What it is for |
|---|---|
| Recipes | Saved sets of steps for jobs you do again and again. |
| Skills | Specific abilities your agents can use. |
| Variants | Different versions of a setup, so you can compare them. |
| Experiments | Structured tests that compare options fairly. |
| Evolution | A record of what improved over time. |
Recipes #
A recipe is a plan for a job, kept so it can be repeated. You do not write them — C5 writes them for you, from work that went well. When a task finishes with a strong result, C5 distils the sequence of agents that produced it; when the same sequence proves itself across several tasks, that shape is kept too.
The kinds of job this tends to produce a recipe for:
- A weekly report that always pulls from the same places
- A review checklist you apply to every document
- A set-up routine you run at the start of each project
You choose when to use one. Pick a recipe in Chat when you create a task, in the task editor on the board, or put one on a timer with Schedules. If what you have written closely matches a task an existing recipe came from, C5 will point that recipe out beside the picker — as a suggestion you can take or ignore, never as a decision made for you.
Each recipe is then judged on results. C5 compares the runs it planned against what similar tasks score when planned from scratch, and shows the difference on this tab and in Analytics. A recipe that turns out to do worse than planning from scratch stops being suggested. One that has not been used enough to say shows nothing rather than a score, because no evidence is not the same as a bad result.
Skills #
A skill is a specific thing an agent knows how to do. Skills are how you teach C5 something particular to your work — the way your team formats reports, the steps your process requires, the rules a certain client expects.
You can turn skills on and off. An agent only uses the ones that are switched on.
Because a skill is instructions your agents follow, every skill is scanned before it can be used, and one with a serious finding is switched off automatically. See Skill safety for what is checked and what is not.
Comparing options #
When you are not sure which of two approaches is better, do not guess. Set them up as variants and let C5 run both, then look at the results side by side.
This is more reliable than trying one, feeling good about it, and moving on. Small differences are hard to see by eye but easy to see in a comparison.
Note Comparisons need a few runs before they mean anything. One good result can be luck.
Automated promotion #
Once your system reaches 100 completed tasks, C5 can automatically promote top-performing prompt variants into production using learned reward models. See Self-improvement milestones for how this milestone unlocks.
Should you use Labs? #
Not at first. Get comfortable with Chat, Productivity, and Library before you come here.
Labs becomes useful once you notice yourself repeating work, or once you want to improve results rather than just get them.
Where to go next #
- Budgets & costs — spending limits and self-improvement milestones.
- Schedules — run a recipe automatically.
- Analytics — see whether a change actually helped.
