OpenAI‘s latest Enterprise Signals report shows enterprise AI shifting from assistance to execution, with frontier firms (top 10% of usage) generating 8.

3× as many output tokens per active user as typical firms, up from 2.6× in January.

This widening gap reflects a deeper operating shift: leading firms connect agents to company context and tools, delegate more substantive work, and make successful workflows repeatable.

The article profiles three startups that embody this pattern.

Basis, which builds AI agents for accounting firms, turns first-day onboarding from two hours into 30 minutes by encoding a reusable skill in Codex that configures accounts and answers questions.

Clay, a go-to-market platform, gives each account a persistent workspace with a dedicated subagent that reviews primary sources nightly and produces daily priorities, saving an hour of inbox triage.

Exa Labs, a web search API for AI agents, turns opportunity discovery into a defined Codex workflow: monitoring repositories, gathering context, creating pull requests, and running tests, with human review before shipping.

Together, these examples show a progression: teach a stable process, provide persistent context, and carry opportunities into tested action.

The article then distills six steps for enterprise leaders: (1) choose one consequential value surface with repeatable, measurable stakes; (2) define the outcome, owner, KPIs, baseline, and guardrails; (3) write the agent’s job description covering triggers, outcome, context, tools, permissions, and review points; (4) build the human system around the agent with named owners for business outcome, domain logic, access, and adoption; (5) make experimentation visible and reusable by capturing what works and packaging it as skills or shared workspaces; (6) carry the operating pattern forward to the next value surface.

The report notes that six months after adoption, early-career employees send 13 more messages per week than executives, indicating room for bottom-up experimentation.