Thousands of IBM Consultants Are Taking a Certification Exam · Issue #205

Opening

On August 14th, IBM announced a strategic partnership with OpenAI. The plan is to integrate GPT-5.6, Codex, and ChatGPT Work into the IBM Consulting Advantage platform, and to push into finance, government, telecom, and retail—as well as into corporate finance, procurement, customer operations, and HR functions.

What caught my eye in the press release wasn’t the technology. It was this line: IBM is building a dedicated OpenAI practice and getting thousands of its consultants and engineers certified as experts through OpenAI’s partner network. Through this collaboration, IBM has also been elevated to OpenAI’s “elite partner” tier.¹

Reader, why do these two sentences matter? Let me cut to the conclusion. What OpenAI is now spreading through the enterprise market isn’t the model—it’s certified people. The IBM deal is simply that strategy playing out at the very top of the market.

IBM Built a Certification Organization, Not a Product Organization

IBM made three promises in this announcement: turning legacy operations into AI-executable workflows, accelerating application modernization and development, and bolting on security and AI risk management. So far, this is what every large SI always says.

What’s new is the shape of the organization tasked with delivering it. IBM has stood up a dedicated unit called the “OpenAI Practice” and defined that organization’s identity as a group of certification holders. It’s not the first time a consulting firm has had its people collectively earn a vendor’s technical certifications. What’s different this time is that the certification isn’t a side effect — it’s the headline of the partnership.

Andy Baldwin, SVP at IBM Consulting, put it this way:

“The challenge isn’t accessing AI technology. It’s integrating AI safely and at scale within complex enterprise environments.”

I think this single line captures the state of the enterprise AI market in 2026 more precisely than anything else. The models are already good enough. What’s stuck is inside the organization.

The Same Design Is Running One Floor Down, Too

A few days ago, I had the chance to read through the facilitator script for a training session OpenAI runs called “Activator Labs 101.” It’s not aimed at partner firms like IBM — it’s a free program for practitioners working inside client organizations.

The session splits AI champions into three tiers: the executive sponsor who sets direction, the transformation leader who designs company-wide rollout and governance, and the Agent Activator, who redesigns repetitive work for a specific team. This training targets that third tier.

The sentence that defines an Activator stuck with me. What separates an Activator from a power user isn’t skill — it’s scope of responsibility. A power user improves their own work, but an Activator builds workflows that “someone else can use and the organization can sustain.” Which means this person has to consult with the workflow owner, security, legal, and IT, and doesn’t hold final approval authority themselves.

Even the hands-on example is spare. Take one internal intake-request task, and let AI handle only classification and routing recommendations. If information is missing, it’s told to stop rather than guess, and anything customer-facing or hard to reverse gets kicked to a human path. Assignment only executes after the intake lead gives explicit approval.

Attend the session and you get a badge. Build an actual workflow, submit it with evidence, and you get a second one — the “Agent Activator” badge. The champion community already has roughly 8,000 members.²

One floor up, thousands of IBM consultants are earning professional certifications. One floor down, client-side practitioners are collecting badges. It’s the same design. Instead of just selling the model, OpenAI is minting, on both sides at once, the people who will take responsibility for running that model inside organizations.

Fragment reviewed - no corrections needed.

Already Using It, But Nobody’s Bought It

Why go this far? The numbers explain it.

The MIT NANDA research team’s “State of AI in Business 2025” became famous for one line: “95% of generative AI pilots fail to deliver measurable ROI.” The report synthesizes over 300 public case studies, 52 structured interviews, and a survey of 153 senior leaders. It’s not a peer-reviewed academic paper, and the sample isn’t large, so it’s hard to generalize this figure at face value.

What I think matters far more in this report isn’t the 95% — it’s a different pair of numbers.

90% of companies have employees using personal AI tools for work, while only 40% have purchased an official company-wide subscription.

That 50-point gap is what’s known as Shadow AI³. And this is OpenAI’s real problem — and its real opportunity.

OpenAI already has a service with 900 million weekly users. Its business customers number over 1 million, and ChatGPT for Work seats⁴ have surpassed 7 million. In April 2026, enterprise revenue exceeded 40% of the total, and the company said it expects it to match consumer revenue by year-end.

This makes clear where the bottleneck actually is. It’s not awareness. It’s not performance. Individual employees are already using it. What’s stuck is the last stretch — the one that converts individual usage into a company-wide contract. To clear that stretch, someone inside the organization has to be able to say:

“I’m going to change this workflow this way. Here’s the security review we went through, here’s where the approval gate sits, and if something goes wrong, I own it.”

The organization’s AI stalls because no one is there to say that sentence. IBM’s certified consultants and the client-side activators are, in the end, exactly the people saying that sentence on someone’s behalf.

The Original Playbook Is Salesforce’s, from 20 Years Ago

If this structure feels familiar, you’re right. It’s what Salesforce has been doing for nearly 20 years with Trailhead⁵.

Salesforce didn’t sell its product — it sold credentials. It taught people for free, handed out badges, and made those badges worth something in the job market. Then something strange happened. Even when a company had no reason to adopt Salesforce, the individual now had a reason to learn it and push for its adoption — because it had become their own career asset.

According to IDC estimates, this ecosystem is projected to generate a net gain of 11.6 million jobs and $2.02 trillion in revenue between 2022 and 2028. That’s not Salesforce’s own revenue — it’s the number generated around it. Of course, this is a vendor-commissioned estimate, sensitive to its underlying assumptions, so it should be read with that caveat.

OpenAI is walking the same path, only much faster. When it launched its first certification program in December 2025, it set a target of certifying 10 million Americans by 2030, and lined up Walmart, John Deere, Lowe’s, BCG, and Accenture as early pilot partners. Coursera and Credly handle the actual issuance. Where Salesforce built up from the community level, OpenAI is starting by driving an anchor straight into the job market. Today’s IBM deal is what happens when you attach a distribution channel — a major systems integrator — onto that anchor.

Models Can Be Swapped, but People Can’t

There’s a counterargument worth raising here. “Still, this doesn’t create as much lock-in as Salesforce, does it?”

That’s a fair point. Salesforce was structured so that switching costs grew the more customization piled up. What OpenAI teaches, by contrast, produces workflow design documents — records of what information to use, what to prohibit, and where a human needs to approve. In principle, these can be carried over to a different model without modification. Requirements specifications are portable.

ibm That’s exactly why certification matters.

When technical lock-in is weak, what’s left is people. The person recognized inside a company as the owner of an AI workflow. The person who has internalized that methodology through hands-on practice. The person who has negotiated with security and legal teams using that vocabulary. Which tool will this person propose expanding next quarter? The one designed in the language they were trained in.

The moment thousands of IBM consultants get OpenAI-certified, the default reference architecture they bring to client companies is already set. And it’s no coincidence that the reference implementation is ChatGPT Work. This feature, released in July 2026, connects Slack, Teams, Gmail, Drive, Salesforce, and SharePoint to autonomously carry out multi-step tasks, requiring human approval before sensitive actions. The principle that Activator training repeatedly emphasizes — “AI prepares and recommends, but a human approves” — is implemented directly as a product feature.

It’s not that the training comes first and the product comes later. It’s closer to distributing the product’s user manual in the form of methodology and certification.

Oz’s Lens

I used to build and run Notion’s Korean community. I learned one thing from that job. A community starts working as a distribution strategy not when people like the product, but when being good at the product becomes a person’s social status. Once someone who’s great at building Notion templates starts getting called “the person who’s good at that” inside a company, the spread stops being driven by the company and starts being driven by that person.

I saw the same pattern over and over while designing GTM strategy at Gamma. Ten product demos moving is slower than one internal champion emerging.

So I think OpenAI’s strategy here is well designed. But there are two things companies adopting it need to be careful about.

First, the Activator role is prone to accumulating responsibility without authority. The training script itself says as much: the Activator coordinates and recommends, while actual change authority sits with the workflow owner and governance lead. On paper, that’s balanced. But in practice, if the company doesn’t properly design the approval gates and exception paths, the one person left holding the bag when something goes wrong is the single practitioner who built the workflow. I see this pattern constantly in consulting. Organizations tend to change the work while leaving the responsibility structure untouched.

Second, a certified advisor is not a neutral advisor. IBM has long held the position that “we’re model-neutral.” An elite partner tier and a dedicated practice is a decision that shifts that position, at least a little. I’m not saying that’s bad. But when you receive an architecture proposal from an organization like that, you need to separate, at least once, whether it’s a technical judgment or a byproduct of the partnership structure.

If you’re at a Korean company, add one more question to this. The major domestic system integrators (SIs) will soon bring similar dedicated teams and certification programs of their own. When they do, the question to ask isn’t “how many people got certified” — it’s “who decides where in our workflow a human has to approve.”

Closing

Here’s the three-line summary:

What IBM signed with OpenAI is a technology agreement, but it’s also a workforce certification agreement. Thousands of credential-holders are the distribution channel.
OpenAI is running the same design upstairs (partner certification) and downstairs (internal Activator badges). It has correctly identified that the bottleneck isn’t performance — it’s who’s accountable.
Workflow design documents can be ported to other models. So the real lock-in isn’t the technology — it’s the person who has internalized that methodology.

If you’re leading internal AI adoption or fielding an SI proposal, I’d suggest checking one question at your next meeting: “Is there a name in the document for who’s accountable when this workflow goes wrong?” If there isn’t, you’re not at the adoption stage yet — you’re still at the experimentation stage.

💬 Have you ever been the one pushing for internal AI adoption, or received a proposal that led with vendor certifications? Tell me in the comments where the balance between authority and accountability broke down the most. I’ll pick this up again in the next issue.

💬 Share your thoughts or experiences on this topic in the comments · 📨 Send this along to a colleague wrestling with internal AI adoption

References & Further Reading

Primary sources

IBM Newsroom, “IBM partners with OpenAI to accelerate secure AI deployment for enterprises across core operations,” August 13, 2026. Link ··· This is where today’s piece begins. Read the paragraphs on “OpenAI practices” and certification before the technical integration items.
OpenAI Academy, “Champions” community. Link ··· This is where the activator program actually runs. You can check the scale for yourself.
OpenAI, “Launching our first OpenAI Certifications courses,” December 9, 2025. Link ··· The 10 million by 2030 certification target and the list of pilot partners are here.
OpenAI, “The next phase of enterprise AI,” April 8, 2026. Link ··· Covers enterprise revenue crossing 40% of the mix and the consulting-partner strategy. Good background reading for the IBM story.
MIT NANDA, The GenAI Divide: State of AI in Business 2025, 2025. Link ··· Look past the 95% figure to the 90% vs. 40% gap — that’s the core evidence behind today’s piece.

Background

OpenAI, “1 million business customers: the fastest-growing business platform in history,” November 5, 2025. Link ··· The source for the 1 million business customers and 7 million seats figures.
Salesforce/IDC, “Salesforce Economy will create 11.6M jobs and $2.02T in revenues between 2022 and 2028,” September 11, 2023. Link ··· A precedent showing how large a certification-based ecosystem can grow. Keep in mind this is a vendor-commissioned estimate.
SiliconANGLE, “OpenAI debuts ChatGPT Work, an agentic tool for automating business workflows,” July 9, 2026. Link ··· Worth reading alongside the certification piece to see how training methodology and product features interlock.
Sify, “95% Companies Failing with AI? An MIT NANDA Report Misread by All,” 2025. Link ··· A rebuttal on how the widely-cited 95% figure should actually be read. Worth pairing with the original report.

📝 Glossary

The author, Kwangseob Ahn, is a professor of business administration at Sejong University and lead consultant at OBF (Oswarld Boutique Consulting Firm). He teaches statistics and data analysis, including business data management and business analytics, while leading GTM and AI strategy consulting in the field, designing the seam between technology and business. He has published academic research on a memory architecture for AI dialogue systems (HEMA) and runs Daily Arxiv, a daily curation of global AI papers. He holds a master's from Korea University's Graduate School of Technology Management and a KMBA. He is the author of Homo Brainless: The People Who Outsource Their Thinking.

Footnotes

Elite Partner tier: The top rank in OpenAI’s partner network. It’s awarded based on the scale of certified personnel, joint go-to-market activity, and enterprise deployment capability. ↩
Champions community: A learning and networking space OpenAI runs for people driving AI adoption inside enterprise customer organizations. As of August 2026, roughly 8,000 people have joined. ↩
Shadow AI: A situation where employees use AI tools for work through personal accounts, without official company approval or contracts. Productivity rises, but risks remain around data leakage and audit trails. ↩
Seat: In enterprise software, one seat means one account, or one user’s license. 7 million seats means 7 million accounts that a company is paying for. ↩
Trailhead: A free online learning platform run by Salesforce. Completing courses earns badges, and Salesforce has built a system where those badges function as real credentials in the job market. ↩