Teams of AI agents that plan, execute, and review entire projects: Gremium.ai, our own product, built by a small team in under six months

The challenge

Knowledge workers, consultants, and project managers who run research-heavy projects spend most of their time on the same loop: break the work into tasks, research, write, track status, repeat. Each of these steps is work AI could take on, but a chat window is the wrong shape for it - a single conversation cannot hold a multi-week project with dependencies, review steps, and real deliverables.

Our thesis for Gremium.ai, transfactor's own product: the right unit of delegation is the whole project, not a single chat message. A user should be able to hand an entire project to a team of AI agents that plans it, works it, and reviews its own output, while the human decides how much oversight to keep - per project, not as a one-size-fits-all setting. The internal target we set for this audience: up to 80% reduction in manual work.

The hard part is not getting an AI to answer a question. It is coordination: breaking work down, managing what depends on what, reviewing quality, recovering stuck work, and giving people enough control to trust the system in the first place.

What we built

Gremium.ai is a workspace where a user creates a project and hands it to a team of AI agents. A planning agent breaks the project into tasks and works out which tasks depend on which, so deep research happens first and report writing comes last. Execution agents then work the tasks with 20 specialized tools - web search, deep research, reading files, writing documents, creating charts - and review agents quality-check every result before it counts as done.

The outputs are real deliverables, not chat transcripts. A built-in report writer plans the structure of a report, fills it in chapter by chapter, adds real charts, and runs a final editing pass. Finished documents are then turned automatically into presentation decks and shareable web versions. Reports can be shared with outside stakeholders, who can ask questions and get answers grounded only in the report itself.

All of this is visible as it happens. The workspace shows tasks moving across a board, a live map of how tasks depend on each other, and what each agent is doing right now. A built-in assistant can also manage projects directly on the user's behalf.

How we built it

Every task moves through a defined series of stages, from planned through in progress and review to completed, so the system always knows exactly where each piece of work stands. Separate services handle planning, execution, and review independently, and a safeguard watches for stuck work and restarts it. The system is built to run many projects in parallel and grow with demand.

Humans stay in control through a per-project setting with three levels of agent independence. At the lower levels, agents pause and ask a person before making judgment calls - and before interrupting anyone, an agent can first consult a more capable AI to see if the question truly needs a human. At full independence, the agents run the whole project on their own. Access to stored data is strictly controlled throughout, with dedicated security hardening built into the product.

What this build demonstrates

Gremium.ai is a working product, built almost entirely by a small team in under six months - roughly 72,000 lines of code. That is the point: with the right way of coordinating teams of AI agents, a small team can ship a system this deep. Staged work, oversight that can be dialed up or down, and full visibility into what agents are doing are the same patterns we bring to corporate innovation engagements where AI needs to do real work under real oversight.