Every large IT services firm has run an agentic AI reorg since last summer. Most ran the same one: cut the experienced people, keep the org chart. One of the largest IT services firms said in July 2025 that it would cut about 12,000 roles, mainly middle and senior management, in what it called a “future-ready transformation.” A global consulting firm set aside $865 million in September 2025 to exit staff it could not retrain.
| Phase | Where the agent lives | Who gains | Who loses |
|---|---|---|---|
| 2026 | Inside each application | Incumbents with distribution | Point tools with thin data |
| 2027–2028 | In a harness across systems | Data platforms, frontier labs, harness vendors | Screen-heavy SaaS, seat licensing |
| 2029–2030 | On the record, under controls | Ledgers, charts, loan files, ontologies | Dashboard vendors |
| Beyond 2030 | In a governed loop | Whoever holds data, harness and enforcement together | Anyone selling a screen |
The model is not the moat. The moat is the data, the controls and the record. So an agentic AI reorg that sheds the people who know the ledger and the loan file gives away the one asset the table says will still be worth money.
The matrix was built for the 2026 row
Most large firms run a matrix. Industry verticals such as banking, insurance and retail own the client and sell. Horizontal service lines such as application services, testing, infrastructure and data own the people and deliver. That shape fits the first row of the table. There, the agent lives inside each application and the vendor ships it. The vertical sells the rollout, and the horizontal staffs it.
It stops fitting in the second row. A leading analyst firm said on July 1, 2026 that up to $234 billion of enterprise application spend is at risk from agents by 2030. That is about 20 percent of enterprise application SaaS spending. One of its managing vice presidents said agents bypass the application itself. So “this breaks the link between user growth and revenue growth,” he said. The same break hits a service line priced by the head. A horizontal built to staff a screen has nothing to staff once the agent works across systems instead.
The cuts hit the people who hold the record
The firms are not cutting at random. Rather, they are cutting the middle and the top. A frontier lab’s economic index report for June 2026 found a gap by experience. Workers with 15 or more years put the share of their tasks that AI can already do about 10 points lower than first-year workers do. Read one way, that is resistance. Read the other way, though, it is the part of the job that is still human. Knowing what a complete loan file looks like, or which ledger entry is wrong, is the third row of the table. Still, that knowledge is walking out with a severance letter.
One large firm is at least pointing the other way. Its CEO wrote in a business magazine that providers must move from systems integrator to “AI builder.” They must underwrite outcomes such as claims-processing speed, he wrote. And they must leave the pyramid for people who combine business and technology skills. He gave loan origination at a top-ten bank as the kind of target he means.
More on agentic AI and the IT services business
- The Diamond, Not the Pyramid: My June post on how agents remove the junior base and what replaces the pyramid.
- The Question Clients Will Ask: Why every IT services firm will soon be asked what it sells when the hours go away.
- How Fortune 1000 Firms Organize: The operating model the buyers are building, which the sellers need to mirror.
- Govern the Harness: Who owns the harness across the Five-Stage Roadmap, and why it is not the model.
- Fifty Enterprise Cases: What 50 deployments say about where the value actually landed.
What I would do with the org chart
Follow the agent. I would organize the firm around where the agent lives in each row, not around the service line that used to sell the hours. That gives me four moves. None of them starts with a layoff list.
Keep the verticals, re-charter them
Today a banking vertical owns a client list and a sales target. Instead, it should own the record: the loan file ontology, the ledger semantics and the controls a regulator expects around them. Then its sellable asset is the record it can hand an agent on day one, not the people it can put on a floor. That is the 2029 row, and it is where the margin goes.
Collapse the horizontals into two practices
A harness practice builds and runs the loop across systems: tools, MCP servers, memory, evals and the agentic AI SDLC that ships it. Application services, testing and infrastructure fold into it. A controls practice owns enforcement: agent identity, credential scoping, audit trail, autonomy tiers and the kill switch. I lay those out in my post on the six layers of agentic AI architecture. Risk, compliance and process operations fold into that one. So one practice makes the agent capable, and the other makes it governable.
Deliver in pods, not projects
A pod has a record owner from the vertical, a harness engineer, a controls engineer and an agent supervisor. Together they own one outcome and the P&L for it. A frontier lab’s study of agent autonomy in practice found that 73 percent of agent tool calls still have a human in the loop. In a pod, that human is the supervisor, and the supervisor is the person with 20 years in the domain. In my work in regulated loan origination, the one person who could tell an agent what a complete file was had two decades in underwriting. Not a certificate in prompting.
Retrain the experienced people to specify, not to code
The consulting firm’s chief executive is right that most of them will not become engineers. She is wrong about what the firm needs them to become. The scarce skill in the last two rows is writing down intent, the definition of done and the eval cases an agent must pass. That is an intent.md and a test set, and a senior underwriter writes a better one than a senior developer. So the agentic AI reorg I would run moves those people into the record and the controls, because that is where the moat sits.
Here is the same table, with the unit that should own each row.
| Phase | Where the agent lives | Who owns it in the firm | What the firm sells |
|---|---|---|---|
| 2026 | Inside each application | Vertical plus application practice | Rollouts, by the seat |
| 2027–2028 | In a harness across systems | Harness practice | The harness, by subscription |
| 2029–2030 | On the record, under controls | Vertical as record owner plus controls practice | Outcomes, by the result |
| Beyond 2030 | In a governed loop | Outcome pods with a P&L | A share of the loop |
Finally, sequence it by row, not all at once. This year, stand up the harness practice out of application services. Next year, the controls practice out of risk and operations. By 2029, pods with their own P&L. Any agentic AI reorg that starts with the severance list instead of this table is selling the moat to pay for the restructuring.
What does your firm’s org chart look like once the hours go away?
Was this post useful? One click tells me what to write next.
Subscribe to the Blog: I publish a new post every weekday on one agentic AI governance question. Enter your email and each post arrives in your inbox the moment it goes live.
For more on this topic, please check out my three books on agentic AI for regulated enterprises:
- Intent In, Outcomes Out: the thesis, why intent in, outcomes out replaces instructions in, results out.
- Earned Autonomy: the method, the EAI-1000 framework that scores how much autonomy an agent has earned.
- The Governed Loop: the architecture, the six layers that decide outcomes, govern the harness, not the model.
© Dr. Harish Kotadia, Ph.D., All Rights Reserved, 2026
Dr. Harish Kotadia, Ph.D., is an Enterprise AI Architect with 20+ years of IT consulting experience serving Fortune 100 clients, specializing in agentic AI systems built on Anthropic Claude, AWS Bedrock, and Google Vertex AI.
Disclaimer: This blog post is based on publicly available academic publications, vendor documentation, open standards, and news items from reputed media sources linked above. This post is intended for educational purposes, to help the enterprise agentic AI community build a shared vocabulary from public, authoritative sources.
Views and opinions expressed here are my own and do not represent those of any employer or client, past or present. The analysis presented is my independent interpretation of the published sources linked above and does not constitute legal, financial, or consulting advice of any kind.