Devin Desktop

A dated reading of what is claimed, reported, and independently verified in the current evidence.

Current reading

The readiness gap, in one scan

AI-assisted assembly · derived results

Evidence strength Critical

Decision

Human editorial judgment · 2026-06-28

Proceed with caution

Why: Strong primary-vendor maturity (Rust rewrite, 30% token efficiency, Nubank 8-12x/20x) and real ACP openness, but the multi-agent core is research-preview, subagents are ~40% reliable, and the review-risk profile on sensitive code is explicit. Aligns with Rule #29 caution on payment/financial-adjacent code.
Next: Scoped 2-week pilot on a non-sensitive codebase (no auth/payments/PII paths), measuring useful first-draft rate vs the stated 70%+ target and weekly hours saved before any broader rollout; migrate any Cascade-dependent automation to Devin Local before July 1, 2026.

Constraints

No named blocker is present in the current public projection.

Evidence summary

AI-assisted assembly

Counts and dates only. Raw signals, private excerpts, trust records, and internal corpus material are not published here.

Publication record

Initial public reading

This is the initial public reading. No earlier readiness change is recorded.