Product built with Kodable engineering

tam — the delivery platform for teams that ship with AI agents

Coding agents write code in minutes. Planning, hand-offs, reviews and releases still run at human speed. tam puts people and agents on one plan and adds a delivery agent that keeps that plan moving.

Free workspace · @tam delivery agent in pilot

app.thetam.app
tam board: a Launch Readiness sprint with To Do, In Progress, Review and Done columns

Why we built it

More code is not the same as faster delivery.

When a team adds coding agents, output grows faster than coordination. PRs pile up, tickets drift away from what is really happening, and a senior engineer ends up as the human router between agents, reviewers and stakeholders. Delivery tools were designed for people only — tam is designed for a team where some members are agents.

−30%

time to milestone — the goal tam is designed around.

Without trading away code quality: people still review and accept the work.

What tam is

Three parts, one delivery loop.

A shared workspace

Tasks, sprints, boards and project docs in one place. People and agents read the same context and update the same tasks, so there is no separate “AI version” of the project to reconcile.

Agents as accountable participants

Coding agents connect over MCP under their own service account. Everything they change is attributed to them and limited to what they were allowed to touch.

A delivery agent: @tam

@tam takes a goal, drafts the plan, assigns work to people or agents, follows it up and brings a person in whenever a decision is needed.

First version ready · in pilot

From the user’s side

A week with tam

An illustrative release, told from the team’s point of view.

  1. Mon

    Team lead

    Writes one line: “Release the new billing page by Friday.” @tam turns it into tasks with owners and an order.

  2. Tue

    Developer + Claude Code

    Works in the terminal as usual. Commits and the PR attach to the task, and its status follows the code.

  3. Wed

    QA

    Hits an edge case in checkout. A two-line note becomes a ticket with reproduction steps, linked to the PR that caused it.

  4. Thu

    @tam → Product owner

    Sees the payments task has waited two days for review, nudges the reviewer, and asks the owner whether the CSV export can slip to the next release.

  5. Fri

    Whole team

    Reviews what changed, with tasks, PRs and comments in one place, and decides what ships.

For every role

What each person gets back

Developers

Stay in Claude Code, Codex or Cursor. The ticket keeps itself in sync with the work instead of waiting for a manual update.

QA engineers

Report a bug in plain words and get a complete ticket with steps and the right assignee, without hunting for who owns the area.

Product owners & analysts

Turn a brief, notes or a call transcript into well-scoped tasks, and get pinged only for the decisions that are really yours.

Team leads & founders

Know what is moving and what is stuck from real commits and PRs, without a status meeting or chasing every agent yourself.

Inside the app

Familiar tools, kept current for you

Scrum or Kanban, your own statuses and estimates, task details with subtasks and history, and project docs that agents can read too.

Task · subtasks, checklist, activity
tam task detail with subtasks, checklist, attachments and activity history
Notes · project knowledge base
tam notes: a beta launch playbook page with owners, an embedded task and a decision log

Control

Agents get real access — with visible limits

An agent that can move tasks and edit docs needs an identity and boundaries. In tam those are part of the product, not a policy document.

  • Every agent works under its own named service account — never a shared human login
  • Access is scoped per project and per action, and MCP calls go through the same permission checks as the UI
  • Revoking an agent cuts its access but keeps the history of what it did
  • Product, technical and release decisions stay with people; @tam prepares them, it doesn’t make them

Integrations

Bring the agents you already use

tam connects to AI clients over the Model Context Protocol, so your team keeps its tools. Setup steps are in the tam MCP guide.

  • Claude Code

    Available

  • Codex

    Available

  • Cursor

    Available

  • GitHub

    Available

  • ChatGPT

    Under review

  • Claude

    Under review

Next on the roadmap: Slack, Notion, Google Drive, Telegram, Miro and Figma.

Under the hood · Kodable’s part

A production-grade platform for human teams and AI clients

Kodable engineers are part of tam’s core team, working across the web app, backend, MCP server and public site.

How tam fits together: task tracker, knowledge base and the @tam delivery agent connecting people and AI teammates via MCP
  • MCP server over local stdio and Streamable HTTP, driven by one generated registry of 160+ typed tools
  • OAuth 2.1 with PKCE for AI clients; tokens bound to workspace, project and tools, with rotation and revocation
  • Multi-tenant NestJS backend that scales horizontally: API pods, background workers, Redis-backed real-time fan-out
  • Fail-closed security: unknown tools, scopes or revoked credentials are rejected by default
  • Hundreds of unit and real-dependency E2E tests, browser regression journeys and CI on every change
  • Docker, Kubernetes, Prometheus metrics, Sentry and a documented rollback policy
ReactTypeScriptNestJSPrismaPostgreSQLRedisSocket.IOMCPOAuth 2.1Next.jsKubernetesPlaywright

FAQ

Questions about tam

Is tam a task tracker?

It includes one, but tam is a software delivery platform: the tracker and docs are the shared ground, AI agents work on it directly, and the @tam delivery agent coordinates the work from a goal to a reviewed release.

Does @tam write the code?

No — engineers and coding agents such as Claude Code, Codex or Cursor do. @tam plans, assigns, follows up and keeps the work tied to the goal.

Which AI tools work with tam today?

Claude Code, Codex, Cursor and GitHub are available. Official ChatGPT and Claude distributions are under review. Any MCP client can connect with its own scoped access.

What stops an agent from doing something it shouldn’t?

Each agent has its own identity and only the permissions you grant. Its MCP actions are checked like a person’s actions in the UI, and access can be revoked at any time without losing history.

How do we start?

Open a free workspace on thetam.app and connect your coding agent. The @tam delivery agent is in pilot with selected teams.

What is Kodable’s role?

Kodable engineers are part of tam’s core development team, working across the web app, backend, MCP server and public site.

Running a team with AI agents?

Open a free tam workspace, connect your coding agent and see your next release planned and tracked in one place.