Team pitch

2026-10-22

The software factory

Automated tests and live monitoring. Our agents fix what they find.

Proposal

The software factory

Fimo watches itself. Our own coding agents do the work. People decide.

FIMO-2988 · the plan Operating model Admin mockup

The software factory

Fimo finds its own problems. Our agents fix them. You decide.

  1. 01FindTests and live checks run on a schedule.
  2. 02FixA coding agent opens a PR with the fix.
  3. 03DecideYou review it and merge.
Every morning: what needs youFimo admin · Today

Today

We find problems by chance, and we follow up by hand

Bugs come in faster than we can sort them, and after each release a person must remember to look again.

95bugs filed in 30 days
19still waiting in Triage
6release checks done by hand

How it works

Checks find, agents fix, people decide

Fimo finds the problem. A coding agent picks it up, finds the cause and opens a PR. A person comes in only when a decision is needed.

flowchart TB
  C[Checks<br/>run on a schedule] --> F[Finding]
  F --> A[Coding agent<br/>picks up the finding]
  A --> P[Draft PR]
  P --> V[Verifying<br/>a watch guards it]
  V --> D[Closed]
  A -. needs a decision .-> Y[You<br/>Slack ping]
  Y --> P
  V -. fires again .-> F
Every change goes through a PR. After the fix ships, Fimo keeps watching for a few days.

Checks

Tests and live checks watch the whole product

Tests can read live data from every project and find patterns that no person could see by hand.

  • A routine asks the same question twice
  • An agent runs but never finishes
  • SEO is not ranking your website
All checks in one placeFimo admin
How a check works

Two kinds of rule. Both are small pieces of our own code.

SQL rule A query on our own data

Fast and almost free. Runs every morning on every project.

-- routines that spend credits but find nothing
select routine_id, count(*)
from runs
where credits > 0 and findings = 0
  and started_at > now() - interval '7 days'
group by routine_id
having count(*) >= 5
Jev A model reads what the agent did

For what a query cannot see. Reads a sample of runs and answers one question.

Did the agent ask the person something
they had already answered?

→ Run 4812: yes. "Which locale?" was
  answered on 12 Oct.
Why we build them ourselves
  1. It knows FimoA generic tool sees crashes. Our checks see product bugs: a routine that forgets, a run that costs too much.
  2. Each check is codeIt lives in our repo, has tests, and is reviewed like any other change.
  3. Agents keep them currentWhen a feature ships or a check is too noisy, an agent opens a PR to add or fix it.

Before and after

A routine that forgets its instructions, told twice

Some routines ignore their own instructions. We saw it by chance. With hundreds of routines on many projects, nobody can find this by hand.

TodayFIMO-2850 · real dates
  1. Routines ignore their instructions. Customers get the wrong reports.
  2. Someone sees it by chance, while testing something else.
  3. The bug is filed.
  4. Still in Triage. Nobody knows how many routines it hits.
8 daysfound by chance, still not fixed
With the factoryhow it would go
  1. A check sees runs that ignore their instructions. It opens a finding with the routines it hits.
  2. A coding agent picks it up and finds the cause.
  3. It opens a draft PR, with a failing test first.
  4. You review and merge. A watch checks the routines for 7 days.
1 morningfound at 06:00, about 15 minutes of a person

A finding

A fix is done only when a watch says so

Every finding shows what happens, the proof, and what comes next. After the fix ships, a watch reads the same signal for 7 days.

A finding in VerifyingFimo admin
Findings · VerifyingFimo admin

Your day

One digest, rare pings, PRs to review

The agent tells you what is wrong, in a few lines. You get one digest each morning, and a ping only when a decision needs you.

09:00 · every day

Fimo MonitoringAPP09:00

Fimo today: 2 need you

SEO titles lost again on 7 of 34 blog sites. Claude is on it. → finding

AI search costs $0.91 a run (+44%). Decide: accept or cap? → finding

Checkout fix closed · Nightly E2E green · 3 dependency bumps merged

Chart · routine cost, 14 days

only when a person must act

Fimo MonitoringAPP11:42

Preview stays blank after you come back

@JS Claude found the cause and opened a draft PR. It needs your review.

Video · 8 s of the blank preview
Review PROpen finding
Fimo MonitoringAPPthread reply

Fix verified: 7 of 7 clean days. Finding closed.

Where this goes

Monitoring is step one of the software factory

Each step gives the agents more autonomy. We add a step only when the one before it works well. People review every change until we trust it.

  1. We start hereMonitorChecks find the problem.
  2. NextTriageAn agent finds the cause.
  3. NextCreate PRAn agent writes the fix and a test.
  4. LaterReviewPeople review. Agents help.
  5. LaterMergeSafe fixes merge by themselves.

More detail: the plan · the operating model

Source peek

Open at head
Path copied