One year of Unblocked at Tradeshift: context engineering for AI coding, and phenomenal adoption far beyond engineering

Here is a story every CTO knows. Your complex Technology stack includes a couple of rarely used legacy PHP apps, and your last PHP developer resigns – ten years in the company, the only person who truly knew that remote, legacy corner of the platform. The week after, a critical defect lands on exactly that code. Or a security finding. Or a tech-debt item, because of course it is. The engineer you assign is talented but has never touched PHP, and there is nobody left to ask. 

Historically, companies panic at this point: switch to stress mode, outsource the maintenance, accept a long ramp-up, higher risks and pay twice.

We built a better answer – or rather, in May 2025, we selected one. And we were very, very wise.

A knowledgeable assistant with permanent, all-time memory

Unblocked is a context engine that has read everything we have and is continuously learning new stuff: our GitHub repositories, Jira, Confluence, Google Drive, Slack and our websites. 

Think of it as a phenomenally knowledgeable colleague with permanent memory – one who can read the code and explain the undocumented areas with diagrams, links and pointers. Ask why the invoice dispatcher behaves the way it does, and it answers from the code itself, the pull request that changed it, and the Slack thread where we debated it years ago! 

Our non-PHP developer? She asks, she gets the map, she fixes the defect. Better, she even comes up with the plan to AI refactor that legacy PHP code into Javascript, closing a tech debt item. Unblocked delivered understanding when she asked questions and gave her AI agents context so they could execute the refactor with her babysitting them.  No panic, no outsourcing, no waiting for an expert to wake up in another timezone. Unblocked is 100% available, 24/7, and never tired. No more work held hostage by a question hanging in Slack or in a Jira comment.

Adoption is phenomenal – and not just in Engineering

Since rollout, Tradeshifters and their agents have asked more than 50,000 questions, from basic to very advanced questions, growing from roughly 500 a month in spring 2025 to about 7,000 a month in June 2026, with ~75 weekly active users. My favorite detail: the busiest Slack channel for Unblocked is #ask-product – ahead of #ask-engineering – with #compliance, #ask-integrations and #dev-ops right behind. Ops, integrations, services and customer success teams ask complex, technical product questions and get self-sourced accurate answers on the spot. That is knowledge democratized.

The results show up where it counts

In our core defect project, we multiplied our defect fix rate by 4x, we are now closing lesser priority items that had been open for a year or more, precisely the undocumented or low priority ones nobody wanted to touch. 

Unblocked’s code review has screened so far hundreds of pull requests and flagged over 50% of them for code logic improvements, from minor tweaks to error-or-crash-or-hang prevention and a few flagged for security concerns.

It is also how we manage change in teams. When we ramped a brand-new vendor Pod, the data told a clear story: Unblocked usage led to an accelerated delivery output by about a month faster! 

New employees and vendors self-serve their ramp-up instead of queueing on our senior engineers. That is autonomy, measured.

The 2026 twist: an AI asking an AI

In 2026, we recommended Claude for AI coding to our Engineers, retired Copilot, and every Tradeshift engineer now has access to Claude Code and Unblocked. Since February, Claude connects to Unblocked through its MCP server or CLI – and this changed the game again: over 5,000 MCP agent tool invocations already, peaking near 800 a week in June, with equal split between Claude Desktop and Claude Code/CLI using AI Coding agents.

Because here is the truth about context engineering: an AI asking another AI for context is far better than a person typing keywords into an AI Chatbox.  Claude AI asks sharper questions to Unblocked, in higher volume, with zero ego – and it writes more accurate code on existing code, legacy refactoring and defect fixing because it actually knows our institutional history and the code. Retrieval-augmented context is what turns a great coding model into a great colleague.

1 year in

Products change. Teams change. People move on, and their context used to leave with them. 

Unblocked is how Tradeshift keeps that institutional memory of the technology stack and apps together. Providing the accuracy and the efficiency we need while everything keeps moving at accelerating rates. 

I said it a few weeks after going live with Unblocked in 2025 and I will say it again in 2026: we were very, very wise to select Unblocked for our needs at Tradeshift.