AI agents are starting to take on whole company workflows!

Hey, Shii-chan here! Today's story isn't about a new product feature — it's about how companies are actually putting AI agents to work, and I wanted to share it.

OpenAI News openai.com

What was announced?

OpenAI News shared case studies from three companies — Basis, Clay, and Exa Labs — on how they're using AI agents to change the way they operate. Each one has built agents into a different part of the business: employee onboarding, sales account management, and developer integration work.

Why it matters

The article points out that "frontier companies" — the top 10% of AI users — are now generating 2.6x to 8.3x more output tokens than they were back in January 2026. That's not just "using AI more"; it points to a shift from one-off Q&A toward actually handing over parts of a workflow to AI.

What changes

Here's what each company is doing:

• Basis (onboarding): cut first-day onboarding for new hires from two hours to thirty minutes. An agent runs the integrated setup following company-specific steps, freeing up the HR team to focus on culture and support instead
• Clay (sales account management): deal information that used to be scattered across the CRM, email, and Slack now gets pulled together overnight by a dedicated sub-agent per customer. In the morning, a coordinating agent lists out the day's priorities, cutting down the roughly hour a night sales reps used to spend sorting email
• Exa Labs (developer integrations): to push toward "Exa everywhere" for its search API, an agent handles spotting integration opportunities, gathering context, opening pull requests, running tests, and putting together a weekly report — leaving humans to focus on choosing opportunities, making judgment calls, and handling outside relationships

Dive Deep

Based on these examples, the article also lays out steps for bringing agents into a workflow:

• Pick one workflow with clear strategic priorities and measurable outcomes
• Define success metrics — cycle time, quality, cost — and who owns them
• Write the agent's "job description": triggers, permissions, required evidence, and stop conditions
• Set up who decides on staffing, access control, rollout, and day-to-day operation
• Make experiments visible and turn them into something reusable
• Carry successful patterns forward into the next project

What stands out across all these steps is that none of them jump straight to full automation — they draw a clear line between what's handed to the agent and what stays with a human, and expand from there.

Wrap-up

• Output token generation at frontier companies is up 2.6x to 8.3x compared to a typical company
• Basis cut onboarding time from two hours to thirty minutes
• Clay uses a per-customer sub-agent to streamline sales account management
• Exa Labs hands developer integration work to agents to push "Exa everywhere"
• A useful set of examples if you're a leader or manager thinking about how to bring AI agents into your own company's workflows!