How the AI team works

TalkIDE is organized around two teams of named agents rather than a single chatbot: a Business team, led by Vera, and an IT team, led by Mara. This page explains the reasoning behind that choice, how the two teams work together, and what the model means in practice for your projects.

Why a team, not one model?

A single large language model can write code, draft designs, and reason about infrastructure. But it has to do all of those things in the same context window, at the same time, to the same standard. For short tasks that is fine. For a project that grows over weeks, with dozens of decisions layered on top of each other, it degrades quickly.

Role-bound agents address this in a few ways:

• Each specialist loads only the context relevant to its domain, so it can be deeper and more accurate in that area
• Responsibility is visible: you can see which agent did what and why, which makes it easier to identify problems and ask for corrections
• Each team’s lead (Vera or Mara) can route a task to the most appropriate model for it, rather than using the same heavyweight model for every sub-task
• The team metaphor maps to how most people already think about building a product, which makes the product easier to use

This is a deliberate trade-off. The setup is more complex than a single chatbot. It is slower for trivial requests. But for projects that need to stay coherent over time, the structure pays off.

How the two teams work together

Every new project starts in a conversation with Vera by default. Describe your idea, and her Business team turns it into a project brief and a clickable mockup you can try out before any code exists. When you’re happy with the plan, Vera hands everything off to Mara in one click, and the IT team picks up from there without you repeating yourself. If you’d rather skip planning, you can switch straight to Mara and describe what you want directly.

Within each team, the lead — Vera for Business, Mara for IT — is who you talk to. When you send a message, the lead:

1. Classifies the request (is this a design question, a code change, an infrastructure task, or a clarification?)
2. Decides which of their own specialists to involve, and in what order
3. Hands each specialist a brief with the relevant context
4. Waits for their outputs, checks for conflicts, and assembles a response

From your perspective, you mostly see the lead’s replies — Vera’s or Mara’s. The specialist work happens in parallel or in sequence, depending on the task, and gets credited by name inside the lead’s messages when relevant.

Role boundaries and handoffs

Agents do not overlap by accident, within a team or between them. The boundaries are enforced by the system, not just by instruction. Iris cannot deploy; Kai cannot rewrite a feature’s requirements; nobody on Vera’s team writes the application itself. This makes it easier to reason about what changed and why, and it limits the blast radius when something goes wrong.

Handoffs are explicit, both within a team and between the two teams. When Iris finishes scoping a feature, she produces structured requirement notes for Theo rather than a free-form description. When Nia finds a failure, she files a structured report that Theo can act on directly. When Vera’s team finishes a plan and a mockup, she hands it all to Mara at once — the brief, the decisions, the mockup — so the IT team builds from it without you having to re-explain anything.

What the model does not do

The team metaphor is useful, but it has limits worth naming:

• The agents are not human. Neither Vera nor Mara get tired or bored. Theo does not take shortcuts out of laziness. But they do make mistakes, and when they do, you correct them the same way you would correct anyone: by describing what went wrong and what you want instead.
• The team does not make decisions behind your back. Every choice that matters is surfaced to you before it is acted on.

If the team gets stuck on something that requires human judgment at the infrastructure or security level, that is what Concierge is for.