GitLab Duo

The trait you will feel every day is that nothing has to be copied anywhere. An issue becomes a merge request without leaving the tab it was filed in, and the same assistant is in the VS Code and JetBrains extensions when you go back to writing. Adjacency is worth more than raw capability once you use a tool eight times a day.

What you do not get is a reason to be here if you are not here already. Pick it when GitLab is the system of record. Pick GitHub Copilot when your work lives on the other host.

The risk is blast radius. Terminal execution and repository writes are both recorded on this row, and no container boundary of its own is. Whatever confinement exists is inherited from the execution context, which in a DevOps platform is the context that already reaches production. A prompt that goes wrong here does not stop at a working copy.

What it does right is naming things. Work is expressed as flows built from a published catalog of agents rather than one open-ended assistant, so what a run is allowed to attempt is declared in advance and can be argued about before it executes.

1. Capability is assembled rather than monolithic. Planner, Data Analyst and Security Analyst are separate documented units combined into flows, which makes the division of labour inspectable in a way a single system prompt is not. 2. The same definition runs from three entry points, so behaviour should not depend on where it was launched.

2. Nothing is published on evaluation. No benchmark, no measurement of whether decomposition beats a single pass, and no stated criterion for a completed flow. The architecture is legible; its effectiveness is asserted.

Read the packaging, not the demo. The base tier ships inside subscriptions customers already renew, which means the company is spending margin to make the seat harder to leave rather than chasing a new line item. That is the correct move for an incumbent with distribution and no model of its own, and it is why pricing power here belongs to the subscription rather than to the agent.

Moat: the repository, the pipeline and the contract, in that order. Position: safe to build on if you are already a customer, and worthless as a reason to become one.

Procurement is half solved and half opaque. Access control needs no new vendor because it is the same account my team already administers, and runs execute on CI/CD runners we already budget, so there is no separate compute line. That is the easy half.

The hard half is the meter. Duo Pro and Duo Enterprise are per-seat add-ons without a published figure, and platform usage consumes Credits bought separately, so I cannot hand finance a number for sixty people before a call with sales. Approved with conditions: a quoted annual cap, and Credits alerting wired to my dashboard, not theirs.

The interesting part is that it is an MCP server as well as a client, so my own tooling can drive it instead of only feeding it. That is the rarer half of the protocol and most vendors skip it. Self-hosted weights are real too, through a gateway I install myself, which is the only path here that keeps inference on hardware I control.

The catch is that the path exists only on the Self-Managed deployment, and the code itself stays shut. I can route it, I cannot read it. Grudging respect for the gateway.