OpenAI Opened Its Agents API to Everyone This Week. What UK SMEs Can Build Now.

OpenAI opened its Agents API to all developers this week, removing the enterprise-only restriction that previously limited who could build durable, multi-step AI agents. The same week, HubSpot reported that 19% of its Pro Plus customers were using Aviator agents by August 2026 - double the rate from earlier in the year - with monthly AI actions up 3.5x. The combination of falling access barriers and rising mainstream adoption marks a clear shift in what is practically available to a 15-person professional services firm or a 30-person logistics business.

What is the OpenAI Agents API and why does this week's change matter?

Until this week, building durable multi-step AI agents with the OpenAI platform required enterprise-tier access. The updated API is now open to any developer. The key capabilities now broadly available include:

• Durable sessions - the agent remembers context across multiple steps without restarting
• Tool use - the agent can call external APIs, query databases, and interact with your existing software
• Subagent coordination - one agent can delegate steps to a second specialist agent and then resume with the result
• Flexible execution - you can run agents on OpenAI's hosted infrastructure or on your own servers

For UK SMEs, this matters because the ecosystem of third-party developers who build integrations for tools like HubSpot, Xero, and Slack can now offer agent capabilities to SME customers. You do not need to build anything yourself.

What does a durable session actually do?

Without durable sessions, each AI interaction started from scratch. With them, an agent can pause mid-workflow and resume when conditions are met - including waiting for a human to approve a decision before continuing. That is the human-in-the-loop pattern that makes agents safe to deploy in customer-facing or financial workflows.

A practical example: a lead qualification agent receives an enquiry from a website form. It researches the company, scores the lead against your criteria, drafts a personalised first-contact email, and then pauses. A notification appears in your Slack with the draft email. You click approve. The agent sends the email, logs the interaction in your CRM, and schedules a follow-up task in three days. The whole process - from form submission to draft in Slack - takes under 90 seconds.

Three UK SME workflows that this unlocks

Workflow	What the agent does	Human touch-point
Lead qualification	Reads enquiry, researches company, scores lead, drafts follow-up email	Approves or edits email before send
Invoice exception management	Monitors aged debtors, drafts chase emails at 7/14/21 days, escalates disputes	Reviews escalated disputes only
Job completion monitoring	Reads field service data, flags missed SLAs, creates daily Slack digest	Receives digest, acts on flagged items

Each of these workflows follows the same principle: the agent handles the volume, a person handles the judgment calls.

What adoption looks like now

HubSpot's August 2026 data gives a concrete benchmark. In a mainstream B2B SaaS product with a large SME customer base, 19% of Pro Plus users were running AI agents. That figure was half what it is today earlier in the year. The rate of adoption is accelerating, not plateauing. Separately, Anthropic disclosed that 30,000 internal agents processed over 1 billion decisions in August 2026, with real-time monitoring blocking roughly 0.002% of those decisions - demonstrating that production-grade agent deployments at volume are now demonstrably safe and reliable.

What do you need before you start?

Three things:

1. At least one business tool with an API or webhook - most modern SaaS tools (HubSpot, Xero, Slack, Teams, Salesforce, QuickBooks) have both
2. A clear definition of the first workflow you want to run - one that has a high volume of similar tasks, a defined input, and a clear output
3. A human approval step built into any workflow that touches customers or finances

The SMEAutomate deployment approach begins with a single workflow. Once that is working and the team trusts it, additional workflows add incrementally. The standard 7-14 day deployment window covers scoping, build, testing, and handover for the first agent.

What about the tools you already pay for?

The point of the Agents API becoming broadly available is not that you need to start using OpenAI directly. It is that the tools you already use will accelerate their own agent capabilities. HubSpot's Aviator, Xero's AI layer, and equivalent products in Sage and QuickBooks are all building on the same underlying infrastructure. Choosing an AI agent deployment partner that works across those tools - rather than building a standalone custom solution - means you benefit from every upstream improvement without additional work.