Demos don’t count. Deployments do.
AgentKit gives teams a path to production…less glue, clearer roles, faster wiring.
What happened
OpenAI launched AgentKit…a toolkit to build, deploy, and operate AI agents in real workflows (support, sales, ops).
Source: https://openai.com/index/introducing-agentkit/
Why it matters
Roadmaps don’t move on slides. They move when a workflow ships in production and stays healthy. AgentKit cuts the plumbing so teams can focus on the job to be done.
Define “done” before you build
Support: the ticket is auto-resolved or routed, handled in five to seven minutes, with customer satisfaction around 4.3–4.6 out of 5.
Sales: the lead is enriched and routed, sales cycle time gets shorter than last quarter, and meeting rates improve over your baseline.
Ops: the task is created, owned, and completed on time, with service levels hit about 95 percent of the time.
14-day pilot plan (keep it simple)
Pick one workflow (support, sales, or ops).
Write the “done” line in one sentence.
Ship the smallest version to production.
Measure two things only: time to live and cost to serve.
Review on day 14. If it can’t prove both, pause or kill it.
Guardrails (day one)
Audit logging on.
Redaction/scope for sensitive data.
Human-in-the-loop on first decisions.
Rollback path documented.