Generative ai update: enterprises move past pilots “The era of AI pilots is over, the era of AI operations has begun,” said Manjeet Rege, director of the University of St.

Thomas Center for Applied Artificial Intelligence, in comments featured by Tech Target and surfaced on the university’s news site on January 16, 2026. The framing is blunt: leaders aren’t asking whether to test generative models anymore; they’re asking how to run them for actual users without the wheels coming off. ‘The pilot era is over’: enterprises flip the switch Rege ties the shift to a pragmatic set of themes—moving from toy demos to integrated systems, folding agent-driven workflows into existing tools, and tightening human oversight.

AI pilots era over: He sketches the change in boardroom conversations this way: “Gone are the days when executives ask, “Should we try this?

Should we test this?” Now they’re asking, “How do we run this reliably for hundreds of users?”” He adds a simple thesis line: “The era of AI pilots is over, the era of AI operations has begun,” Tech Target’s piece, highlighted by the University of St. Thomas on the same day, attributes several specific priorities to this new phase: agent-driven workflows, stronger human oversight, and “responsible, secure, cost-effective” deployment.

The list reads like a reaction to the last two years—lots of experimentation, plenty of surprises, and mounting bills. Whether those priorities land as checklists or hard constraints will depend on the messy details of implementation, which the summary doesn’t enumerate.

Source: University of St.

Thomas news post on Rege’s Tech Target feature.

Adoption spikes from 5% to 80% as 2026 nears Gartner’s forecast sets the backdrop. Going into 2026, more than 80% of enterprises will have tested or deployed applications infused with generative AI, up from less than 5% in 2023. The jump tracks with the timing of ChatGPT’s public debut in November 2022, which kicked off a wave of trials and proofs-of-concept that rarely needed much internal selling. More than 80% will have tested or deployed by 2026 (Gartner).

Less than 5% had tested or deployed in 2023 (Gartner).ChatGPT landed in November 2022. One caveat embedded in Gartner’s language: “tested or deployed” lumps together everything from short-lived experiments to shipping software.It signals broad intent, not uniform depth.

Rege’s message about moving to operations implies a higher bar—uptime, security reviews, access controls, audit logs, and change management.

The summary also says return on investment remains a worry for decision-makers. That tracks with the last year of belt-tightening across tech budgets, but no numbers were shared on what “acceptable” ROI looks like or how it’s measured across different use cases. Agents, oversight, and the ROI test Rege contrasts the early phase—standalone chatbots and demo sandboxes—with a next phase built around agents tucked inside existing systems.

His description: “GenAI is dissolving into the enterprise stack.
Users will access it through ERP forms and CRM workflows, supply chain screens, ticketing systems and not by going (to a GenAI tool.
It is becoming like electricity.

You don’t see it, you use it. You use what’s built on top of it.” The appeal is obvious: people keep working where they already work, while LLMs draft text, summarize records, or kick off follow-up tasks behind the scenes. Embedding agents in live workflows, though, changes the failure modes. A bad suggestion in a chat window is one thing; an overconfident agent pushing a change into an ERP screen is another. That’s where Rege’s call for “stronger human oversight” meets reality—review steps, clear rollback paths, and unambiguous logs of what the model saw and did.

There’s also the cost side.

The themes include “cost-effective” deployment, which suggests a step beyond open-ended experimentation. Fine-tuning, retrieval pipelines, token limits, and caching strategies sound dry until the invoice arrives.

None of the material linked here breaks down those costs or points to comparative benchmarks, so it’s hard to judge whether the shift to embedded agents reduces spend or just moves it to different line items. The ROI reminder is doing a lot of work. Trade-offs to keep in view: Chat UI familiarity vs. Embedded convenience: centralize oversight or spread it across dozens of screens. Speed of automation vs.

Human-in-the-loop controls: fewer clicks vs. Higher assurance. Experimentation freedom vs. Cost discipline: quicker iteration vs — predictable bills.

None of these are new tensions in software, but the stakes feel higher with systems that confidently hallucinate. Vendor risk also tilts differently when text generation is entangled with ticketing and CRM flows instead of living in a separate sandbox. Observability, red-teaming, and policy audits have to follow the work into each integrated touchpoint. From ChatGPT’s debut to enterprise normalization The maps cleanly to the quotes and stats: November 2022: ChatGPT’s launch ignites mass experimentation. Teams spin up prompts, compare models, and collect early wins and surprises.

2023: Pilots proliferate.

Lots of chatbot trials, document summarizers, and code assistants run on exception paths rather than core workflows. 2024–2025: Guardrails harden and costs get scrutinized.
Security reviews, data governance, and platform choices start to gate what ships. Going into 2026: Operational rollouts become the rallying cry. Rege’s line about “hundreds of users” sets the scale target, not just a dozen power-users in a lab.
Rege’s take paints generative models as plumbing under familiar screens rather than a destination app. If that’s how 2026 plays out, the interesting action won’t be in a chat window at all. It’ll be in whether agent steps are visible enough to audit, whether teams can quantify the value beyond anecdotes, and whether the cost curves flatten with smarter architectures and caching instead of ballooning as usage grows.

The quotes capture the mood shift, but the hard questions (what’s the bill, what’s the benefit, what’s the blast radius) still need line-item answers.

The maps cleanly to the quotes and stats:

November 2022: ChatGPT’s launch ignites mass experimentation. Teams spin up prompts, compare models, and collect early wins and surprises.
2023: Pilots proliferate. Lots of chatbot trials, document summarizers, and code assistants run on exception paths rather than core workflows.
2024–2025: Guardrails harden and costs get scrutinized. Security reviews, data governance, and platform choices start to gate what ships.
Going into 2026: Operational rollouts become the rallying cry. Rege’s line about “hundreds of users” sets the scale target, not just a dozen power-users in a lab.

Rege’s take paints generative models as plumbing under familiar screens rather than a destination app. If that’s how 2026 plays out, the interesting action won’t be in a chat window at all. It’ll be in whether agent steps are visible enough to audit, whether teams can quantify the value beyond anecdotes, and whether the cost curves flatten with smarter architectures and caching instead of ballooning as usage grows.

On the evidence presented here—a Tech Target feature amplified by the University of St. Thomas and a set of Gartner figures—the generative ai update looks less like a hype cycle and more like a set of unglamorous implementation chores: integrate, supervise, secure, and prove it was worth doing. That’s not a dismissal; it’s the work. The quotes capture the mood shift, but the hard questions (what’s the bill, what’s the benefit, what’s the blast radius) still need line-item answers. More details at AI pilots era over. More details at enterprise AI operations.

Generative ai update: enterprises move past pilots “The era of AI pilots is over, the era of AI operations has begun,” said Manjeet Rege, director of the University of St.

AI pilots era over: He sketches the change in boardroom conversations this way: “Gone are the days when executives ask, “Should we try this?

The list reads like a reaction to the last two years—lots of experimentation, plenty of surprises, and mounting bills. Whether those priorities land as checklists or hard constraints will depend on the messy details of implementation, which the summary doesn’t enumerate.

Source: University of St.

Thomas news post on Rege’s Tech Target feature.

Rege’s message about moving to operations implies a higher bar—uptime, security reviews, access controls, audit logs, and change management.

His description: “GenAI is dissolving into the enterprise stack.
Users will access it through ERP forms and CRM workflows, supply chain screens, ticketing systems and not by going (to a GenAI tool.
It is becoming like electricity.

There’s also the cost side.

None of the material linked here breaks down those costs or points to comparative benchmarks, so it’s hard to judge whether the shift to embedded agents reduces spend or just moves it to different line items. The ROI reminder is doing a lot of work. Trade-offs to keep in view: Chat UI familiarity vs. Embedded convenience: centralize oversight or spread it across dozens of screens. Speed of automation vs.

Human-in-the-loop controls: fewer clicks vs. Higher assurance. Experimentation freedom vs. Cost discipline: quicker iteration vs — predictable bills.

2023: Pilots proliferate.

Lots of chatbot trials, document summarizers, and code assistants run on exception paths rather than core workflows. 2024–2025: Guardrails harden and costs get scrutinized.
Security reviews, data governance, and platform choices start to gate what ships. Going into 2026: Operational rollouts become the rallying cry. Rege’s line about “hundreds of users” sets the scale target, not just a dozen power-users in a lab.
Rege’s take paints generative models as plumbing under familiar screens rather than a destination app. If that’s how 2026 plays out, the interesting action won’t be in a chat window at all. It’ll be in whether agent steps are visible enough to audit, whether teams can quantify the value beyond anecdotes, and whether the cost curves flatten with smarter architectures and caching instead of ballooning as usage grows.

The quotes capture the mood shift, but the hard questions (what’s the bill, what’s the benefit, what’s the blast radius) still need line-item answers.

The maps cleanly to the quotes and stats:

November 2022: ChatGPT’s launch ignites mass experimentation. Teams spin up prompts, compare models, and collect early wins and surprises.
2023: Pilots proliferate. Lots of chatbot trials, document summarizers, and code assistants run on exception paths rather than core workflows.
2024–2025: Guardrails harden and costs get scrutinized. Security reviews, data governance, and platform choices start to gate what ships.
Going into 2026: Operational rollouts become the rallying cry. Rege’s line about “hundreds of users” sets the scale target, not just a dozen power-users in a lab.

AI pilots era over: enterprises switch to operations

Generative ai update: enterprises move past pilots “The era of AI pilots is over, the era of AI operations has begun,” said Manjeet Rege, director of the University of St.

AI pilots era over: He sketches the change in boardroom conversations this way: “Gone are the days when executives ask, “Should we try this?

Rege’s message about moving to operations implies a higher bar—uptime, security reviews, access controls, audit logs, and change management.

There’s also the cost side.

The quotes capture the mood shift, but the hard questions (what’s the bill, what’s the benefit, what’s the blast radius) still need line-item answers.

AI pilots era over: enterprises switch to operations

Generative ai update: enterprises move past pilots “The era of AI pilots is over, the era of AI operations has begun,” said Manjeet Rege, director of the University of St.

AI pilots era over: He sketches the change in boardroom conversations this way: “Gone are the days when executives ask, “Should we try this?

Rege’s message about moving to operations implies a higher bar—uptime, security reviews, access controls, audit logs, and change management.

There’s also the cost side.

The quotes capture the mood shift, but the hard questions (what’s the bill, what’s the benefit, what’s the blast radius) still need line-item answers.