What actually shipped in AI tools this summer
A working leader's changelog for June to September 2026 — what changed in Cursor, Claude and ChatGPT, what it means if you run a team, and the parts worth ignoring.
A working leader's changelog for June to September 2026 — what changed in Cursor, Claude and ChatGPT, what it means if you run a team, and the parts worth ignoring.
Cursor Projects puts an agent in charge of other agents. It is a genuine shift in how software gets built — and it moves the hard problem from writing code to specifying…
Cursor’s self-hosted machines move tool execution onto infrastructure you manage. Review what information still leaves the worker and what the agent is allowed to do.
A training cutoff is a warning about built-in knowledge, not a guarantee about every answer. Use current sources and check both the facts and the conclusion.
Read release notes for capabilities, material improvements and operational changes. A practical filter for deciding what deserves investigation.
Review ten common prompt habits through controlled comparisons. Keep useful roles, examples, and constraints, and remove instructions that fail to earn their place.
Claude, ChatGPT and Cursor all have a feature called Projects. They solve different problems, and assuming they are equivalent is how teams end up with the wrong one.
Project instructions can make AI coding work easier to review. What belongs in them, how loading differs between tools, and where real controls are still needed.
Legacy code exposes missing context, weak tests and model limitations. A practical way to make small agent-assisted changes more verifiable.
Cursor lets you pin a skill as a Custom Mode. Use it to make a recurring procedure explicit, with checks for the behavior that still needs verification.