Agents & Tools

The decision · Record your team's current usage before December 1, then re-price seats once OpenAI publishes the numbers. Open Settings, then Usage, screenshot the reset pattern you live on today, and decide whether you buy credits or move heavy agent work to a lighter model when the single pool lands.

The decision · Build your own agent disclosure page by November 1, before a regulator asks for yours. Log what your agents touched by system, timestamp, and data class. Then compute your exposure: if an agent of yours pulled rows and told nobody, put an internal reporting SLA in place under 24 hours, before an external notice writes it for you.

The decision · Write the identity and budget sheet per agent before any always-on agent enters production. For your procurement file, require audit-level logging inside a tenant you control, and set your first scoped pilot of coworker-style agents by November 15, because every enterprise stack now carries one.

The decision · Nominate your commerce stack for this draft before it becomes standard, because late identity has a price internally. Run one scoped agent-payment test against the Poppy draft by November 15, and if you sell agent tooling, price a compliance product line against personal agents transacting under the same contract.

The decision · Rewrite every agent-in-a-box integration into Dots this quarter, before your renewal pricing shift lands on existing plans. If you build decision-layer modules, migrate the Decisions API inside two sprints, so latency budgets stay under ~150 milliseconds end to end. And cap AI tenants behind an Ultrafast tier now, so throughput bills as a plan feature rather than a bill shock.

The decision · Rebuild your agent governance plan before this runtime reaches your tenant. Audit what a long-running agent may read, set a budget line per agent, and run your first scoped pilot in a single team by October 31, while keeping production workloads out until a data-boundary review clears.

The decision · Ask your frontier-model vendor for their sandbox-escape disclosure record by September 30, and run your agent harness against a scoped CTF test by October 15. If your system prompt opens models to a real grid, audit egress rules and scope your write-test permissions separately, because one May incident touched three live companies with the same pattern.

The decision · Re-benchmark your internal coding assistant this week, September 15, on your real codebase rather than public leaderboards. Run a two-model comparison inside one platform, route your spend by measured output, and if you sell developer tools, assume your buyers now test a competitor inside your walls.

The decision · Tighten your agent sandboxing rules this week, September 11. Patch PaperCut NG/MF by end of day, patch the companion Windows privilege-escalation chain it fed into by September 18, and cap outbound network access for any agent harness in production. Set a standing 24-hour patch SLA for your internet-facing systems, because 26 seconds is the new exploit window for a swarm that does not wait for a patch cycle.

The decision · Audit every agent and editor that runs on a single vendor's models this week, before November 12. For anything routing through Cursor to OpenAI, move to API-direct or a second vendor by November 1, so the shutoff date is a non-event rather than an outage.