How Tilt turned AI into a trusted engineering teammate with Tembo

Why Tembo

Tilt evaluated several AI engineering platforms. Some had promising ideas but were not ready for production, with go-live dates several months out. Others had mature products but offered little opportunity for Tilt to influence the roadmap.

Tembo stood out because it delivered both. The product was ready to deploy, and the team worked closely with Tilt throughout the rollout, quickly incorporating feedback into the platform.

“You guys having a working product, but also being small enough that we felt like we could collaborate on the roadmap and see the product grow with us, was definitely a huge win.”

James, Tilt

From deployment to production

Most of the rollout involved deploying Tembo into Tilt's Azure infrastructure, including virtual machines, private networking, and key vaults, all managed with Terraform.

Once that foundation was in place, Tembo itself was running within a couple of days in each subsequent geography where Tilt operates.

Linear became the first integration. Engineers immediately began assigning small tickets to agents while additional repositories and integrations were connected over time.

Results

Cleanup work happened faster. Before Tembo, cleanup work constantly lost priority to feature development. Now those tasks run continuously in cloud environments, generating hundreds of narrowly scoped pull requests that engineers review and merge when they are ready.

One project standardized years of inconsistent telemetry logging across the codebase. By consolidating multiple logging patterns into a single strongly typed approach, AI-generated code became significantly more consistent, following the newer preferred patterns instead of legacy ones. Similar cleanup work standardized Entity Framework configuration across the application. The result has been far fewer nits blocking pull-request reviews.

Instead of occasional large cleanup projects, Tilt now ships a continuous stream of small improvements that would have been difficult to justify or execute manually.

Customer investigations moved beyond engineering. Instead of immediately escalating customer-service questions to engineering, non-engineers can now @mention Tembo in Slack to investigate issues using production logs and database context. Dynamic data masking helps protect personally identifiable information before a developer becomes involved.

Standing up local environments for non-engineers is not practical. Tembo gives the team a secure, codebase-aware way to investigate issues with a significantly lower barrier to entry. For Tilt, that workflow has become just as valuable as AI-assisted coding.

The biggest lesson

The biggest surprise was not code generation. It was identity.

Allowing non-engineers to use MCP servers is a major productivity boost, but it also introduces risk. A user investigating production through a SQL MCP server with elevated access could expose data that their own identity is not permitted to send to AI tools.

Tembo runs in Azure under its own managed identity through Microsoft Entra ID. That gives the platform a defined access boundary rather than inheriting a user's broader permissions. The foundation unlocked entirely new workflows.

“Tembo is like an engineer that's smart enough, but not super dangerous. It has the access it needs to read logs, look at some data, and commit draft PRs, and you can trust it to suss that out for you.”

Kyle, Tilt

Building together

Deploying a self-hosted platform inevitably surfaced edge cases. Through a shared Slack channel, feedback regularly turned into product improvements within days.

Rather than feeling like a traditional vendor relationship, Tilt worked alongside the Tembo team to shape features as they rolled out.

Looking ahead

Tilt sees two major opportunities. The first is planning: the team wants non-engineers to collaborate with agents on implementation plans before work reaches engineering, allowing developers to review the plan before execution begins.

The second is richer cloud environments. Expanded virtual-machine support, including macOS environments for iOS development, together with preconfigured tooling and reusable skills, will allow agents to start with more context and take on increasingly complex engineering work.

In their words

“I view Tembo as an engineer that's dangerous enough to get stuff done, but smart enough to do it right.”

Kyle, Tilt

“The best testimonial I can give is that I got in trouble for too much adoption. I was told to keep the beta group limited, but people kept using it during the beta before we were ready to release it.”

James, Tilt