Codat is an advisory intelligence solution purpose-built for modern commercial banking. Banking teams rely on it to turn raw accounting and ERP data into the types of insights that help them deepen relationships with their customers and grow revenue.
The complexity of that product shows up in the codebase. The team has more than 200 microservices, each with its own repository. And behind each of those services is years of decisions spread across Slack, Confluence, Jira, and Azure DevOps.
To address the challenge of working in such a complex codebase at the speed of AI-native software development, Codat turned to Unblocked.
The challenge: When agents act without context, rework adds up#
Codat’s engineering team has gone all-in on AI-native software development. The DX benchmark for AI-authored code share is 20% – Codat is at 75%. But their priority isn’t just more AI-generated code; it’s ensuring that everyone on the team can ship AI-generated code that adheres to their quality standards without taxing the system.
But agents start every session as strangers. An engineer needs to give them the relevant context to produce mergeable code. For Codat, that context existed across 200+ repositories, years of Slack threads, Confluence pages, Jira tickets, and Azure DevOps boards. "There's a lot of institutional knowledge from past decisions that lives outside our docs." explained Matt Thompson, a staff software engineer.
The Codat team first tried to solve the context problem the obvious way: give agents more information. They hosted an on-premise enterprise search tool, built their own MCP servers, and connected everything they could. But the team quickly realized that access to information wasn’t the problem; it was the understanding that comes with knowing which source was authoritative, how information was related, and what applied to the task at hand.
Without real understanding of scattered context, agents break conventions, cross boundaries, and recreate code that already exists. The faster they work, the faster rework stacks up.The Codat team needed a way to move at AI speed without producing code that took longer to clean up than it saved to generate.
The solution: a context engine for developers and agents#
After getting a recommendation for Unblocked from a friend, Matt quickly started a trial. Seeing Unblocked jump in with the right answers in Slack prompted excitement from the team: "OMG. What is this????”
Unblocked bridged a gap that retrieval tools couldn’t. It connects to the sources where Codat's context already lives and builds a knowledge graph of how Codat's system works, why it's built that way, and who the experts are. When an engineer or an agent asks a question, Unblocked returns a reconciled, cited answer grounded in the team's actual code, past PRs, and conversations.
Every engineer at Codat now uses Unblocked. To keep agents on track, they connect the Unblocked MCP server to Claude Code so it can pull context automatically while it works. And if someone wants a fast, human-readable answer, they ask questions directly in Slack or the web app.
"I came back from a week off and my Unblocked session had expired. I didn't notice before I started working. Very quickly, Claude was doing web searches for Codat, trying to figure out what our products were. I cancelled it, re-authenticated Unblocked, and told Claude to try again. One question, got the answer, carried on." — Matt Thompson, Staff Software Engineer
The results: With a 92 NPS, Unblocked is Codat’s context layer for agents#
The most common question Matt gets from other engineering leaders is whether rules files and a stack of MCP servers can already do what Unblocked does. His answer goes straight to maintenance cost. "The problem with documentation is that it's static. You end up spending your time keeping it current instead of shipping features. Unblocked works off the code, docs, and Slack conversations you already have, so the right context just shows up without any additional overhead."
Codat is seeing Unblocked’s impact in its numbers. PR throughput is up 30% since their last DX survey, and change failure rate has reduced from an already small 4% down to 3.5% in the last three months. Plus, the latest DX snapshot shows a massive 20 percentage point increase on perceived quality. The team credits increased use of Unblocked as a key contributing factor across these metrics. Matt says:
"The value massively outweighs the cost, put it that way." — Matt Thompson, Staff Software Engineer
Codat engineers now reach for Unblocked during planning, code generation, code review, and bug investigation, handing off context gathering so agents can work with the full picture.
"I find myself pretty frequently just saying, 'ask Unblocked about this.’ I know it already has the answer." — Matt Thompson, Staff Software Engineer
