Closing the AI software development context gap and winning Salesforce and Airbnb

Context fragmentation, not model quality, was the real bottleneck blocking AI-speed software development. So we built an AI context-layer platform with tools to refine, govern, and access that context across the SDLC, and it now powers active engagements at Salesforce, Airbnb, Persistent and 8x8.

Every enterprise team had access to the same frontier models, yet agentic workflows were stalling inside real codebases. Talking to developers and PMs at ten enterprises surfaced the same pattern: agents had no shared semantic understanding of the system they were operating on, and humans were spending their day re-stitching context by hand.

If context is the bottleneck, then the unit of value isn't another agent — it's a queryable, refinable, and governable context layer over the enterprise's own code, exposed where engineers and agents already live.

1. 01

   Defined the context layer and the integration-ready tools (MCP, skills, and a CLI), composing with custom SDLC workflows involving various models and IDEs.

2. 02

   Designed a multi-agent orchestration system with specialised agents, model escalation, and graceful degradation for robust enterprise performance without silent failures.

3. 03

   Stood up a semantic eval pipeline — golden datasets, canonical issue model, embedding + LLM-as-judge — and used precision/recall as the weekly product metric.

4. 04

   Shipped a dual GTM motion: VS Code + Chrome extensions for bottom-up developer pull, and a platform sale for top-down enterprise adoption, with discovery mechanics embedded.

The platform achieved 10+ enterprise engagements, including Salesforce, Airbnb, Persistent and 8x8. The eval-driven fixes lifted tool precision from 58% to 87%, and a bottom-up motion across VS Code and Chrome extensions reached 1,000+ developers.

"One thing I took away: your target audience can be right even when your GTM is wrong. Going bottom-up didn't replace the enterprise sale. It got us feedback fast enough that by the time we reached Salesforce and Airbnb, the product was actually ready."