The Control Point: Your Next AI Decision Is a Procurement Decision

Quick answer. In fifteen days, OpenAI, Google and Workday each told corporate customers where their AI agents should live — with them.[1] They are not selling the same product. They are competing for the control point: who picks the model, what an agent may touch, and who gets paid. For a mid-market buyer, AI is now a contracting decision, not a technology one.

Fifteen days, three answers

OpenAI went first on 29 September, letting large customers spend part of their OpenAI budget on other vendors' software. Google followed on 8 October with what it calls a single, universal agent for work.[1] Workday answered on 13 October by opening its HR and finance systems to any company's agents — on condition they obey Workday's rules.[1]

Read by what each asks you to hand over, they come apart quickly.

The prize is the control point

The control point is the place where a company decides which model does a task, what an agent is allowed to touch, and who gets paid. Whoever holds it sets terms for everyone else in the chain.[1]

Who	Owns	What it asks you to hand over
Google	Every layer — chips, cloud, models, Workspace, billing	The gateway every agent passes through
OpenAI	The budget line	Your committed spend, now spendable across 32 partners
Workday	The rules	The permission and approval logic for HR and finance
Anthropic	Neither — supplies all of the above	Nothing; it is betting that being everywhere beats owning the place

What Workday's move actually proves

Google's advantage is structural: when a project hits its spend cap the agent stops, because Google runs the bill, and its Agent Gateway — an AI network firewall — applies one rule to every agent at once, including rival models.[1] Its own release lists identity and policy management, sandboxing and real-time spend caps, with Smart Routing choosing between models.[3] OpenAI's is financial: eligible customers can count part of their commitment toward partners including Salesforce, ServiceNow and Harvey.[1]

Workday's is the least glamorous and hardest to copy, because it is not a platform claim at all. Its Trusted Agent Platform now gives any MCP-speaking agent headless access to HR and finance data, while every agent works under the same permissions, approvals and audit trails.[2] Customers pick the model: Claude, Gemini, Llama, Mistral, GPT or Nemotron.[2]

The risk it answers is worth quoting: point an agent at an ungoverned copy of HR data and it reads compensation it should never see; connect it through a system-level account and it acts without approval; either way, nobody can prove what it did or why.[2]

"Headless can't mean lawless," said Workday CTO Gabe Monroy.[2] That is the most important sentence in enterprise software this month, and it is not a feature. It is a position on where authority lives.

Why this lands on your desk now

Eighty-four percent of buyers say agent capabilities will influence their 2027 software decisions, and a third call it a primary deciding factor.[5]

The pricing model is moving underneath them. Only 9% of buyers consider per-user pricing the fairest model for agents, while pay-as-you-go structures rose from 38% to 57% in a year.[5] When the unit of billing changes, the contract changes, and most procurement functions have not caught up.

The cheap-model trap

The obvious response is to stay model-agnostic and buy the cheapest tokens. That is necessary and no longer sufficient.

Kimi K3 lists at $3 per million input tokens and $15 per million output; Claude Sonnet 5.5 lists at $2 and $10 and scores 56 on Artificial Analysis' intelligence index against Kimi's 44.[1] The budget model is not even the cheaper one per unit of work delivered. OpenAI's own pitch was price-led too: GPT-6.1 Sol delivers near-Astra intelligence at a fifth of its predecessor's token prices.[4]

Cheap only wins when something routes the task to the right model and proves the result — a capability of whichever system holds the rules. That is why the control point matters more than the price list.

Four questions for your next AI contract

1. Who holds the rulebook? If the answer is a vendor's platform rather than your policy, you have outsourced authority, not just software.
2. What does leaving cost? A budget commitment pulls other vendors onto someone's shelf. A gateway polices agents on rival models too. A rulebook is hardest to move because it was hardest to write.[1]
3. Can you route models freely, and see what it cost? Insist on per-task routing and per-agent spend visibility, not a blended monthly invoice.
4. Where does the audit trail land? If the record of who approved what lives in a vendor console you cannot export, you do not own your evidence.

Where this lands in your stack

For a mid-market firm the safest answer is the boring one: keep the rules in the system of record you already govern, and let agents connect under your permissions. That is the argument for consolidating on one platform — and equally for putting an intelligence layer over legacy systems instead of rebuilding them. Either way the rules stay yours and the models stay interchangeable.

Bottom line

You are not buying an agent. You are choosing which permanent layer you will live with: someone's gateway, someone's budget line, or your own rulebook.

Pick the one you can leave. Then make the vendor prove in writing that the permissions and audit trail survive without them.

Related reading: SAP vs Odoo: The AI Architecture Question, The Odoo vs SAP vs Oracle TCO Battle and What Is a Secured and Closed AI System?.

Sources

[1] https://www.forbes.com/sites/guneyyildiz/2026/10/18/every-ai-company-now-sells-an-agent-they-are-not-selling-the-same-thing — Güney Yıldız — Every AI Company Sells An Agent. They Are Not Selling The Same Thing, Forbes, 18 Oct 2026
[2] https://investor.workday.com/news-and-events/press-releases/news-details/2026/Workdays-Trusted-Agent-Platform-Now-Delivers-Headless-Access-to-HR-and-Finance-Letting-Customers-Build-and-Run-Lawful-AI-Agents/default.aspx — Workday — Trusted Agent Platform Delivers Headless Access to HR and Finance, press release, 13 Oct 2026
[3] https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026 — Thomas Kurian — Gemini at Work 2026: Introducing Gemini agent, Google Cloud Blog, 8 Oct 2026
[4] https://openai.com/index/devday-2026-recap — OpenAI — DevDay 2026 Recap, 29 Sep 2026
[5] https://www.prnewswire.com/news-releases/new-g2-research-ai-agents-have-moved-from-pilot-to-production-but-oversight-comes-at-a-cost-302908654.html — G2 — 2026 AI Agents Report press release, PR Newswire, 14 Oct 2026