State of AI for development: June 2026

Three things from June that we think change decisions, and what we are telling clients to do about each.

The best model of the month became unavailable within days

Anthropic’s Fable 5 launched at the top of the main leaderboards, then a US export-control directive forced a global suspension roughly three days later. Access only began returning at the end of the month. OpenAI’s new GPT-5.6 family launched restricted to about 20 government-approved partners.

What we are telling clients: frontier-model access is now entangled with national-security process, which makes it a business-continuity risk no amount of multi-region infrastructure fixes. Model neutrality has become an architecture decision. The routing and abstraction we recommended in May on cost grounds is now also a continuity requirement, because you may not get to choose when you switch.

Open-weight models crossed the coding bar

GLM-5.2 became the first open model past 80 on Terminal-Bench, at roughly a tenth of frontier pricing, and in real head-to-head agentic coding it was slower but cheaper and more robust than the leading closed model. The practical gap between top open weights and the closed frontier has largely closed for coding work.

What we are telling clients: we said this about Kimi K2.5 in January and the case is stronger now. Evaluate an open-weight model for cost-sensitive agentic workloads, via API rather than self-hosting. This is also the continuity hedge for the point above.

The adoption numbers say the copilot was the easy part

June’s analyst data was consistent: 88 per cent of organisations use AI somewhere, but only 23 per cent are scaling agentic AI; nearly half introduced AI without redesigning any workflow; and only 12 per cent of CEOs report both revenue gains and cost reductions from it.

What we are telling clients: the blocker is work design, and it will not close without measurement. Baseline where your people are, redesign the workflows that matter, and measure again.

---

This digest is the public edition. Advisory clients get the full monthly review, a tailored edition for their context and the conversation that goes with it.