AI consultants after DevDay 2026: where the work will be

I write this from two sides. My company has held OpenAI Select Partner status since 24 September. Day to day I deploy agents for clients and run a collective of AI consultants. So I read the DevDay launches with one question in mind. What does the client get ready-made, and what does someone have to set up for them?

What DevDay said about partners, and what it left out

The official DevDay recap does not mention the OpenAI Partner Network. It says nothing about certified consultants or the Forward Deployed Experts program. Partners show up in two places. The first is the 32 software companies in OpenAI Marketplace. The second is the 16 tools where you can use your ChatGPT plan allowance.

There is also one sentence about who deploys agents in large companies. For specialist dots, OpenAI says it starts with focused pilots. OpenAI's own engineers will work with the organizations. They will define the agent's responsibilities, its tools and how people approve its work. The other partner named is Microsoft, with Agent 365. Consulting firms are not in that sentence.

The partner program has been running since June. OpenAI put $150 million behind it. It also aims to train and enable 300,000 certified consultants by the end of 2026. There are three tiers: Select, Advanced and Elite. I described how my company reached the first tier and what that status does not guarantee. DevDay announced nothing about the program. It did change the list of problems clients will bring to consultants.

Five jobs DevDay leaves for companies

1. Rules for dots

A dot is an agent that works around the clock on its own cloud computer. It can reach the apps you connect to it. It follows Custom Rules, which allow an action, require approval, or block it. Someone has to write those rules before the agent gets access to email and the CRM. Someone also has to know that a dot's memory cannot be corrected item by item. It can only be wiped as a whole.

2. A data map for GDPR

A European company first asks where the data goes. After DevDay the answer depends on the feature. In ChatGPT, only the Enterprise and Edu plans keep data in the EU. Space, Pages, Meetings and team tasks are not on the list of data that promise covers. The Agents API supports only US data residency. The client needs one page that shows what may be switched on and what may not. I checked it separately, feature by feature, with a note on participant consent for meeting notes.

3. Model costs

GPT-6.1 Sol costs a fifth of GPT-6 Astra's price for input and output. Ultrafast costs six times the standard rate. Luna is twenty times cheaper than Sol. These gaps change the bill for every agent. The client does not know which model is good enough for its tasks. A consultant who measures the cost of one completed task on the client's data has a ready service.

4. Agents on your own account

The Agents API offers the Codex harness as a service. Bedrock Managed Agents does the same inside AWS. For a client that makes three routes: an agent on OpenAI's API, an agent in AWS, or its own harness. Each has different consequences for data, costs and vendor dependence. Choosing and migrating is a project of its own.

5. Plugins and integrations

ChatGPT plugins can now have their own panel, a view next to the conversation and a file viewer. MCP Events starts an automation when something changes in a connected app. Sites can use plugins and show each visitor their own data. A company with a system that has no ready plugin needs someone to build one. This is the most technical of the five jobs.

Which skills and certifications will help

All five jobs are about permissions, data, quality measurement and cost. Prompts are the smallest part of them. These are the skills of a systems architect who understands how a model behaves.

Certification from OpenAI is narrow today. OpenAI Certified exists, but the documented access is for invited Enterprise and Edu organizations. Since 21 September OpenAI Academy has four learning paths. When you pass a course assessment you get a course badge. OpenAI does not call it a certification. I compared what the Claude, AWS and Microsoft exams and OpenAI's courses confirm, and which work each path fits.

OpenAI's goal is 300,000 certified consultants by the end of the year. An exam confirms knowledge. A client will still ask to see a deployment you have built.

How to start before clients ask

Pick one of the five jobs. Build it on your own example within a week: a rulebook for a dot, a data map for one company, or a simple plugin for a system you know. Write down what you did and what you measured. With a project like that it is easier to answer a client's brief. I described where AI consultants find client work and how to agree the scope of a first pilot.

You can also do it in a group. I run a collective of AI consultants where you prepare for AI vendor partner certifications and, after a profile assessment, may get invited to a project. Today we prepare there for the Claude exams through our partner organization.

I collected the full list of launches, with the keynote timeline, in a guide to everything OpenAI announced at DevDay 2026. Start with what affects your clients, and write down one question you will ask them this week.

Sources and fact-check date

• OpenAI: DevDay 2026 recap
• OpenAI: Introducing the OpenAI Partner Network
• OpenAI: Introducing dots
• OpenAI Marketplace
• OpenAI Academy: new learning paths

As of 29 September 2026.

Frequently asked questions