How Lucanet Established an Automated Context Layer for its Agentic SDLC With Driver

About

Lucanet is an AI-powered CFO solution platform built to help modern finance and tax leaders automate and integrate core financial processes. Backed by prominent investors such as Hg Capital, Lucanet serves more than 6,000 organizations across 50 countries, including Allianz, trivago, and Volkswagen.

---

Challenge

Compiling Agent Context Across Large Legacy Codebases

Like many companies operating in the highly-competitive finance vertical, Lucanet relies on agentic solutions as a force multiplier across its SDLC. But as AI-assisted development became more prominent, Lucanet found that manual context consolidation was becoming increasingly unscalable.

Lucanet was not starting from zero. Teams were already using manual steering documentation to help agents understand the product, codebase structure, technologies in use, and implementation constraints.

While that manual pipeline was successful in producing accurate agentic outputs, maintaining markdown context files across hundreds of repos and millions of lines of code required significant engineering bandwidth. When steering documentation drifted from the underlying code, teams had less confidence that agents were working from the right codebase context and implementation constraints. “Our codebases are constantly evolving, and producing high-quality steering documentation by hand was a real struggle for the teams with large, legacy repos,” shares James Musson, Vice President of Engineering at Lucanet. “The better that documentation is, the better the agents perform.”

The challenge also showed up whenever engineers had to work outside the codebases they knew best. Lucanet’s team was asking how to give agents real context while also helping developers who were new to a codebase, or moving between teams, get up to speed faster. When codebase knowledge lived with individual teams or in manually maintained steering documentation, context stayed siloed. That made onboarding a new codebase and reviewing unfamiliar pull requests slower and more dependent on the people who already knew the system.

During his search for a solution that would automate context management at Lucanet, James discovered Driver, and knew the platform would be the ideal fit.

We were asking ourselves: how do we provide real context for our agents? How do we enable developers who are new or moving between teams to get up to speed quickly? Those were the questions that led us to Driver.

---

Solution

Replacing Manual Curation With Automatically Generated Context

Lucanet worked directly with the Driver team to connect GitHub Enterprise, link repos, and add Driver’s MCP endpoint to Claude Code, Cursor, and their CI/CD pipelines. They then piloted Driver’s context layer with some of Lucanet’s more complex codebases, where manually preparing steering documentation had been especially difficult. Instead of asking teams to produce all context by hand, Driver automatically generated structured documentation that agents could use to better understand the codebase.

Today, Driver sits as the core context layer across Lucanet’s development workflow, spanning 200 engineers.

As code changes, Driver automatically updates architecture overviews, onboarding guides, file-level documentation, dependency graphs, and call-chain documentation, shrinking context rot as the company scales. “I was handed the tool and tested it out with no expectations. Quite fast I discovered that it makes things quicker and more effective,” shares a Lucanet engineer. “Getting the right context into the prompt was far more effective than asking Claude to figure it out on its own.”

Since implementation, Lucanet engineers have continued to apply Driver context to more specific workflows. Lucanet now also uses Driver to help query the codebase and annotate incoming support requests before engineering review. The goal is to give engineers better context sooner and reduce the manual investigation required to understand whether an issue relates to configuration, customer understanding, or the code itself. “I integrated Driver’s MCP into our CI/CD pipelines for pull request risk analysis — checking dependencies, configuration changes, test coverage, file criticality,” shares a Lead Engineer at Lucanet. “During code freezes, when I’m reviewing PRs on codebases I don’t know deeply, it helps me understand what a change is actually doing.”

For developers, cross-repo knowledge silos have also been significantly reduced. Teams working across Lucanet’s products now query implementation patterns from adjacent repositories directly inside their coding environment before writing new code, reducing duplication and keeping conventions consistent across the broader codebase.

The feedback from the teams was that preparing the artifacts you need for agents to be successful was quicker, easier, and more accurate with Driver. The result is we deliver more features, and deliver them more accurately.

---

Results

An Automated Context Layer that Scales Alongside Lucanet’s Codebase

With Driver, Lucanet diminished much of the manual work required to set up steering documentation for complex codebases and reduced the number of iterations needed to give agents usable context. Instead of spending repeated cycles prompting, refining, or asking models to explore the codebase on their own, engineers could get to useful outputs faster. Teams could spend less time preparing context and more time moving work forward, helping them deliver more features and deliver them more accurately.

• Automated context layer across 200 engineers
• Faster onboarding and PR review on unfamiliar codebases
• More accurate agent outputs with fewer prompt iterations

Today, James and his team are continuing to explore how Driver can support additional workflows across engineering and adjacent teams. Product, documentation, and support use cases remain areas of interest, particularly where codebase context can reduce the amount of back-and-forth required to answer technical questions.

One of the reasons I was happy to sign the agreement was the way the Driver team approached the trial period. It feels like a partnership, not just a vendor relationship. They want to understand what we’re doing so they can improve the product — that’s worth a significant amount.