How Integral automated 98% of dependency upgrades with Tembo
Integral uses Tembo to automate dependency upgrades and turn recurring maintenance into reviewable pull requests.
Customer story
About Integral
Integral is an AI-native tax firm for Germany. The company holds its own tax and payroll licenses and runs German bookkeeping, month-end closing, payroll, and tax workflows through AI agents on its own platform, with licensed tax advisors and human clerks reviewing and correcting the output.
That combination matters in a regulated market. Integral is not selling software to accountants; it is the accountant, using software and agents to automate rules-heavy tax work while keeping the licensed advisor of record and human review in the loop.
“Tembo is the best agentic tool to shift the development lifecycle to the left. It gives you a framework, but makes you free to use any AI coding agent you want. It's a frameworkless framework.”
Ali Onur Geven, Integral
The challenge
Like many engineering teams, Integral had accumulated a backlog of dependency upgrades and routine maintenance. The work was necessary, but rarely urgent enough to compete with shipping product, so it continued to pile up.
Over time, maintenance became a tax on engineering. Keeping the codebase current required time and attention that could have been spent building new features.
Why Tembo
Integral evaluated several agentic coding tools, but did not want to standardize on a single coding agent or replace the team's existing workflow.
What stood out about Tembo was its flexibility. Rather than being another coding agent, Tembo became the infrastructure layer around them. Engineers could keep using whichever AI coding agent made the most sense while Tembo handled execution, coordination, and visibility across the team.
The first morning
The team's first workflow was a Sentry-driven cleanup task. Tembo generated 12 small, reviewable pull requests, each in under five minutes.
That result immediately validated the team's approach. Since then, the workflow has remained largely the same: write a clear implementation plan, hand it to Tembo, let agents execute the work in parallel, then have engineers review and merge the results.
Results
Today, Tembo handles roughly 98% of Integral's dependency upgrades agentically.
Instead of allowing maintenance work to accumulate, Integral continuously turns recurring engineering tasks into small, reviewable pull requests. Engineers spend less time keeping the codebase healthy and more time shipping product.
What began with dependency upgrades has expanded into a broader approach for handling routine engineering work. Tasks that once sat in the backlog now move continuously through review, allowing the team to focus on higher-leverage work.
The biggest lesson
The biggest improvement did not come from changing coding agents. It came from writing better plans.
Early on, only 12-16% of agent-generated pull requests were merged. As the team improved how they scoped work and wrote implementation plans, their target merge rate increased to 75-80%.
The team's thinking shifted from asking whether agents can write production-ready code to asking how they can consistently give agents work that agents can succeed at. Success became less about choosing the right model and more about creating repeatable, well-scoped work.
Looking ahead
Today, Integral authors implementation plans in Cursor before handing them to Tembo for execution.
Over time, the team expects planning, execution, review, and collaboration to come together in a single cloud development workflow, allowing engineers to move from idea to merged code without switching between disconnected tools.
Ship more with Tembo
Tembo brings background coding agents to your whole team. Start with one workflow and scale from there, with every change reviewed before it ships.