Concho AI turns enterprise codebases into a knowledge layer for AI agents

Concho AI today introduced its flagship platform, an artificial intelligence platform that understands software development and application work, providing deep semantic understanding and organization that enable developers and AI agents to transform and modernize massive, sprawling codebases into something manageable.

“Its job is to provide an intelligence level that you would normally get from a high-priced consultant or an architect to anyone who wants to know what the code says,” Chief Technology Officer Bruce Henderson told SiliconANGLE in an exclusive interview.

Now that the software industry is stepping into agentic AI, developers are discovering that agents can generate enormous volumes of code. For greenfield development and smaller repositories, they work well. Frontier models and industry-standard context window sizes let them “see” entire projects at a time, enabling them to gather enough data and knowledge to do significant work.

However, Henderson argues, as projects age, as they reach millions of lines of code across several languages and have passed through multiple generations of developers, this can erode the ability for AI agents to work efficiently and accurately. Also, as time passes in any codebase, something even more terrible happens: developers come to dislike writing documentation, or simply neglect to do so. It’s common for older codebases to contain “shadow” code that exists for a purpose but its operation and value may have been lost when developers left the company years ago.

“What you’re looking at is a combination of business-critical system and archaeological marvel,” Henderson explained.

Generative AI has made producing new code inexpensive. Recovering knowledge from embedded and old systems remains considerably harder.

Concho analyzes an application and supplies its findings to an existing AI assistant rather than asking developers to adopt entirely separate tooling or coding agents. It’s a middle “fact layer” that acts as a knowledge graph plugin and lets users explore what an application means, including its architecture and business behavior, rather than inspecting individual files and procedures.

This information is then exposed to tools such as Anthropic PBC’s Claude through the Model Context Protocol. This makes it immediately useful to both coders and business users. It means that anyone can sit down, talk to the AI assistant of their choice and receive intelligent, conventional knowledge about the codebase and business data.

Bringing truth to code by making it easier to read

Under the hood, Concho uses what Henderson calls a “cognitive precompiler” because it performs much of the application discovery and implementation work before a user or agent asks a question. The resulting fact model can return a compact architectural overview or descend into a deep dive, backed by specific, sourced evidence.

In the field, multiple customers are already using Concho to understand their systems and build better apps.

For example, Clearwave, a private-equity-backed medical technology company, uses the product as part of its offerings, allowing it to understand “medical coding,” a complex set of pricing guidelines for procedures that don’t always align with standardizations.

The company also has a small group of seasoned developers who act as mentors for the application. Concho is being used to transfer some of that knowledge to other teams without requiring the experts to consult with them every hour of the day for routine maintenance.

“They’re north of 12 million lines of code in their core offering, and it’s been around for a while,” Henderson said. “It’s been through a number of changes, and it is very tangled and complex.”

Concho also extends beyond code and into business value and purpose. For example, the CEO could use the system to feed accurate data to an AI assistant and get an update on what’s happening in the company, all without bothering the engineering team.

“The engineers are a little less grumpy because they don’t have to sit in all these meetings to try and explain tech’s viewpoint,” Henderson added. “Sales and support and ops are all now working through their tool of choice, which is Claude. They’re working through Claude to talk to the Concho engine and do their first-line triage with the AI.”

Clearwave started with the platform as an engineering tool, but the application understanding model proved useful to executives, product teams, sales, support and operations. Henderson explained that this is because of its deep contextual understanding of the company’s application knowledge surface.

Enterprise AI development is beginning to move beyond first-generation coding copilots and vibe-coding execution, which rely on frontier models to search repositories and assemble context for each task. Emerging platforms are instead building persistent representations of applications, business rules and dependencies.

As codebases continue to sprawl and touch more parts of businesses, and the AI agents that control and orchestrate them under the command of humans work with larger amounts of data, that data needs to be curated and made into something that can act as an accurate foundation of how the code works and what the business does.

Concho’s larger bet is that the models writing the code may matter less over time – becoming essentially the engine switched in and out by developers depending on their needs. It’s the quality of the application knowledge supplied to it that makes or breaks the end result. Humans will continue to decide what must change, while agents perform work, but that requires a clearly understood source of code, patterns and business logic first.

“Concho’s work is to read the source code and build this map of everything, this queryable library that either humans can use, or humans can use through AI, or the AI can use on its own,” Henderson said.

Photo: Freepik