AI is now ordinary in engineering. DORA's 2025 research found 90% of respondents using AI at work, and Stack Overflow's 2025 survey found 84% using or planning to use AI tools in development. Trust has not kept pace: 46% of developers actively distrust the accuracy of AI tools against 33% who trust it, and only 14.1% use AI agents at work daily. That gap is the design brief. Engineers will use an agent on their project board when they can drive it from their own tools and see exactly what it did.
1. The REST API: Your Permissions, Nothing More
Kavanah publishes an OpenAPI specification for its REST API. A workspace admin mints an API key, and every request made with it is authorised as the person who minted it, with their permissions re-checked on every request. Change that person's role or remove them and the key's reach changes at once; revoke it and access stops.
2. The CLI: Scripts and Deterministic Tool Calls
The kvn command line signs in with a device flow that a workspace admin approves in the browser, then drives the API from a terminal or a CI job. It can run one agent tool by name, deterministically, or hand a prompt to the agent and let it choose. A direct tool call is fail-closed: if your action policy says that kind of action needs approval, the call is parked for a person rather than run.
3. MCP: The Board Inside Your Assistant
The Model Context Protocol is an open standard for connecting AI applications to external systems. Kavanah runs an MCP server, so you can add it to Claude as a custom connector at kavanah.ai/api/mcp, sign in, and read and write your workspace from the assistant you already use. The connector authenticates as you, so it sees exactly what you can see.
4. Governance Lives in One Place
However a request arrives, the same controls apply. AI employees see only the projects they are given; safe mode removes risky tools such as deletes and outside email; action policies make chosen categories wait for approval; every action lands in a ledger with undo; and an agent cannot grant itself authority. DORA describes AI as an amplifier of an organisation's existing strengths and weaknesses, which is why the controls belong in the platform rather than in each script.
5. Where to Start
Pick one read-heavy job: a script that asks the agent for a sprint recommendation every Monday and posts both lists to the team, or an assistant that answers what is blocked on this project. Run it for a week, read the ledger, then add one write action at a time. Anything in Kavanah that runs on its own or spends AI credits is off until a workspace turns it on.
Engineers trust tools they can script and inspect. Give the agent an API, a CLI and an MCP server, and put the controls underneath all three. For the full checklist and a dated comparison with Jira, Linear, Asana and monday dev, see our project management software for tech companies page.
