Meet us at Disrupt
Your coding assistant. Better equipped.
For the hard questions.
Keep working with the assistant you know. Give it Mechantic: connected understanding of your codebase, and source evidence it can check.
Codex Claude Code GitHub Copilot
At Disrupt, we’re putting Codex to the test—with and without Mechantic.
Want to try Mechantic? Email us to register your interest. Details & contact.
Not another assistant. A better-equipped one.
01 / The hard questions
Your assistant can already explain a function. The difficult questions ask how an entire operation behaves—across files, callbacks, background work and failure paths.
Closing a window, asking work to stop, releasing resources and receiving a late callback are different events. The answer needs to connect them, not just find a cancellation call.
A request can end while a session or another request continues. Understanding the outcome means following the job’s lifetime, not assuming it ends with a single request.
Creating a record, accepting bytes, validating them and notifying other clients can happen at different stages. A convincing answer must not mistake one stage for all of them.
Examples of the questions behind our challenge. The difficulty is in the connections—not obscure trivia.
02 / Your assistant, with Mechantic
On a hard question, finding individual pieces of code is only the beginning. Your assistant has to work out how they fit together—and what the evidence really establishes.
Bring a question that matters to your work: what a change affects, why an operation fails, or whether existing code can safely do the job.
Mechantic supplies connected codebase understanding and supporting source evidence. Your assistant uses that material to work on your question.
Follow the source references. Test the explanation. Challenge the gaps. The developer stays in control—not a black-box verdict.
The aim: get to a useful, well-supported answer sooner.
Faster only matters if the answer holds up.03 / The Disrupt challenge
At the stand, watch two Codex instances in VS Code tackle the same question about the same source code. The difference: one can use Mechantic.
With Mechantic
Source code access, plus Mechantic’s codebase understanding and source evidence.
Without Mechantic
Source code access, using its usual code-reading and investigation tools.
The judge runs in VS Code with access to the source. Open a completed question to see both answers, why the judge preferred one, and the supporting findings and evidence.
Speed is only one part of the result.
Lower is better on all four measures. Incorrect and unsupported claims are selected findings, not exhaustive defect counts.
Simple questions may tie. That’s fine. The hard questions are the test.
A comparison is evidence about that question and that run—not a promise that every answer will be faster or correct.
TechCrunch Disrupt / San Francisco / October 2026
See the comparison. Inspect the answers. Ask us a question of your own.
What does your assistant keep getting wrong—or take too long to work out?
Explore the challenge
Meet Mechantic
at Disrupt.
Join the beta
We’re inviting developers to beta-test Mechantic with their own projects and coding assistants. Tell us a little about your work and we’ll be in touch.
Email us about the betaOpens a draft in your email app, addressed to John with Nigel copied. Edit it and send when you’re ready. No email app? Copy either contact address and write to us from your webmail.