For founders, AI operators, dev shops and small B2B teams

The 14-day kill rule for agentic growth experiments.

If an AI growth experiment has no kill criteria, it is not an experiment. It is theatre with a calendar.

Agents make it cheap to produce more: more pages, more posts, more lead magnets, more dashboards, more polite internal documents about all of the above. That is useful when the loop is learning. It is dangerous when the loop is just decorating uncertainty.

The fix is not a bigger plan. The fix is a short run with a public artefact, a buyer path and a rule for stopping. Fourteen days is usually enough to see whether a small loop creates any signal worth compounding. Not a final verdict on the company. Just enough to stop pretending that motion equals traction.

A good kill rule is written before the run starts.
If you only decide after the numbers arrive, the experiment will mysteriously become «promising» every time.

The minimum useful setup

One market signalPick a source where buyer pain appears repeatedly: support threads, search queries, community questions, competitor complaints, sales notes or churn reasons.

One paid offerName the buyer, the outcome and the price hypothesis. «AI automation for businesses» is not a paid offer.

One proof assetShip a sample report, teardown, checklist, calculator, benchmark or tiny tool that helps before the sale.

One capture pathMake the next step obvious. No hidden form, no vague «reach out if interested», no funnel that depends on memory.

One factual ledgerTrack cash, committed revenue, qualified opportunities, useful replies, shipped assets and killed assumptions. Do not count vibes.

What to kill

Kill the offer if nobody can understand the outcome. Kill the channel if it attracts the wrong audience. Kill the proof asset if people consume it but cannot see the next step. Kill the automation if it creates work that no buyer asked for.

Do not kill a loop just because it is uncomfortable. Kill it when the evidence says the weakest link is structural and the next two fixes would still not create a buyer path. That distinction matters. Some loops need tightening. Some are just expensive hobbies, even at CHF 0.

A clean 14-day rule

By day 3, the proof asset must be live or usable.
By day 7, at least one distribution path must point to it without spam or deception.
By day 14, there must be either a qualified opportunity, a clear buyer-language learning, or a specific bottleneck worth fixing next.
If none of those exists, archive the loop and reuse only the parts that taught something.

The point is not to be ruthless for the aesthetic of it. The point is to protect attention. Agents can keep a weak idea alive forever if nobody gives them a stop condition.

Signal Foundry uses the same rule on itself: public asset first, no fake proof, no paid tools, no spam, factual ledger. If the loop does not create qualified demand, the offer gets tightened or killed.

Run the loop leak check Read field note 001 Back to Signal Foundry