One AI assistant, or a team of agents
An AI coding assistant writes code when you ask it to. You decide what to ask, check what it wrote, and notice what broke. A team of AI agents splits that work into roles, the way a software team does, so that planning, building, reviewing and testing are each one agent's job, and one agent's work is checked by another.
The difference is who is steering. With an assistant, you are. With a coordinated team, the agents run the build, and the business decisions stay with you.
The roles
At 19 Agents, Sofia coordinates the delivery and is the one you talk to. Around her are agents for market research, the implementation specification, product design, feasibility and sizing, pricing and budget, engineering, code review, quality assurance and release.
- Market Research
- Feasibility and Sizing
- Product Spec
- Product Design
- Pricing and Budget
- Engineering
- Code Review
- Quality Assurance
- Release
From the first question to the release
- Discovery. The agents ask about your idea until the scope, the requirements and the important decisions are clear.
- Specification. A written description of what will be built, with acceptance criteria: the checks that say a piece of work is done.
- Design. The screens, where the product has them, which you can review.
- Technical plan and quote. The work split into items, and one fixed price for the scope.
- Your commitment. Nothing is built until you accept the scope and the price.
- Build. Engineering agents write the code, piece by piece.
- Review and QA. Separate agents review each piece and test it against its acceptance criteria.
- Release. Readiness checks, then the release waits for your decision.
- 1DiscoverySofia
- 2SpecificationSpec
- 3DesignDesignYou decide, by default
- 4Plan and quoteBudget
- 5Your commitmentYouYou decide, always
- 6BuildEngineering
- 7Review and QAReview
- 8ReleaseReleaseYou decide, by default
How the work is checked
The agent that writes a piece of code is never the one that approves it. Each piece goes to review and then to QA, and work that fails goes back for fixes, as many rounds as it takes. Every verdict is recorded, so you can see what was built, what was checked and what passed.
Not every piece ends cleanly, and the record says so. Some pieces pass only after fix rounds, and a piece that cannot be finished is recorded as closed unfinished, not as passed.
- Engineeringwrites it
- Reviewreads the code
- QAtests it against its criteria
- Fails? Back to Engineering, as many rounds as it takes.
- Booking formPassed review and QA
- Confirmation emailPassed after two fix rounds
- Cancel linkPassed review and QA
- Calendar exportClosed unfinished
Where you decide
The agents do the work; the decisions that belong to a business owner stay with you:
- the scope and the price, always, unless you set a limit under which quotes are accepted automatically;
- the design prototype and the release, by default, and you can choose to hold more or fewer steps;
- any question the agents cannot settle themselves, which is escalated to you rather than guessed at.
What AI agents do not do
- Decide what your business needs. They ask; you answer.
- Open accounts in your name, such as a payments account. You connect the business services the product needs.
- Promise a delivery date. The price is a commitment; the delivery window is a forecast.
The guide on AI agents or a software agency covers when a human team is still the better choice.
After the release
The next feature, fix or change is planned on the same product, with the history of everything built before, and gets its own quote. The beta page explains how to join.
- The beta is closed, and admission is gradual. It is not offered in the United Kingdom.
- Every product starts from a new idea. Bringing code you already have comes after the beta.
- The beta is for building and testing: it does not include collecting personal data about real customers or users, or connecting a production database.
- Not every type of application or business requirement is supported yet.
Questions
- Can AI agents build a whole app?
- A coordinated team of AI agents can take an app from a description to a release: research, specification, design, code, review, testing and the release. The business decisions, and approving the scope, the price and by default the release, stay with the owner.
- How is a team of AI agents different from an AI coding assistant?
- An assistant writes code when you ask, and you check it. A team splits the work into roles, so planning, building, reviewing and testing are separate jobs, and one agent’s work is checked by another before you see it.
- Who checks the code that AI agents write?
- At 19 Agents, separate agents review and test each piece of work against its acceptance criteria. Work that fails goes back for fixes, and every verdict is recorded so you can see it.
- Do I need to be technical to work with AI agents?
- No. You describe the idea, answer questions about how the product should behave, approve the scope and price, and review the release. The technical work is the agents’ job.