Product agents: the product owner's AI, it knows what the product does, what is broken, what was decided and why | yaab
AI Agents · Product agents

The product owner's agent.It knows what the product does, what is broken, what was decided and why.

It answers your product, support and sales teams from the code running in production, not from outdated documentation, with the source cited.

What goes wrong

Six ways product knowledge goes wrong.

Next to each one, what the agent does instead.

01

Promises the product cannot keep.

Before you promise a client, the agent shows the settings that save without the effect their name suggests, and what the old manual undersold.

02

Documentation nobody knows is wrong.

The deployed code is the only truth. Every contradiction with the documentation is recorded, with the section to correct.

03

The main branch is not production.

The agent reads every component at the commit running on each server, and dates every rule by that commit.

04

Decisions made twice.

Every decision is kept with its reason. A new one replaces the old without deleting the history.

05

Knowledge that leaves with people.

When someone leaves, their open questions and problems move to the person who can confirm them today. The record keeps the original owner.

06

Failures nobody sees.

For every step, the agent records what happens when it fails. Silent failures go first on your engineers' list.

What you get

The product owner's homework, done and cited.

Answers in plain language, each citing repository, file, line and commit.

The record of the product: initiatives, decisions with their reason, promises, risks, dependencies and open questions, each with an owner.

Known problems by impact, each with an owner, the share of usage it affects and what it takes to fix it.

Findings for your engineers: components running the wrong version and screens that do not exist.

A product manual per release, with its catalogs, built from the facts checked against the code and verified before delivery.

Facts kept current. After a release, facts about changed code are in doubt until checked again.

Where each fact comes from

Every fact has a source. A person decides.

Every fact says how it is known and how sure it is.

Questions agreed first.

The questions the agent must answer are agreed with you before it is built.

Done means cited.

The agent is done when it answers every agreed question with a citation.

Gaps with a name.

What is not known is recorded with the name of who has the answer, never as an estimate.

Who decides.

The agent prepares and answers. The product owner sets the priorities.

Where it applies

Products nobody can hold in their head anymore.

For SaaS teams, and for IT departments.

SaaS products with many integrations

where behavior changes per client platform.

Products after turnover

in the product or engineering team.

Fintech and health software

where what is stored, where and for how long has to be answered with certainty.

Technical due diligence

before buying a company.

Onboarding

of a new product manager or a support team.

Legacy systems

whose builders are gone, as the first step before modernizing them.

Legacy software modernization →

A product agent is the first module of a Company Brain, one connected record of how your company works.

Case in production

A SaaS product whose documentation no longer matched the code.

Nobody could say with certainty what the product did, what was broken or what had been promised. The agent now answers from the code in production, cites the file and line, and names who knows when it does not.

Code in productionCited answersDecision recordManual per release
  • Every fact marked with where it comes from and how sure it is.
  • Contradictions found between the documentation and the code.
  • Key rules checked again independently, surfacing faults nobody had recorded.
  • A product manual rebuilt from the verified facts with every release.

Bring us the product nobody fully understands anymore.

One working session, 45 minutes: we pick twenty questions your product team cannot answer with certainty today, and map how to answer them from production code.

We reply within one business day.