TL;DR
- OpenAI's data shows enterprise agent usage in research workflows surged 56x between November 2025 and June 2026 — this is not a pilot, it's a production reality.
- Engineering use cases grew 27x in the same period, signalling deep technical adoption across industries.
- Real deployments — like a financial-sector rollout that saved 1,200 hours per month for 700 users — prove the ROI story is concrete and repeatable.
- The window for "evaluating AI" has closed. Swiss IT organisations that aren't deploying are already falling behind.
The Experiment Is Over. Did You Get the Memo?
There's a certain comfort in calling something a "pilot." It buys time, it manages expectations, and it lets you nod thoughtfully at conferences while describing your organisation's "AI journey." But sometime between late 2025 and mid-2026, the rest of the enterprise world quietly stopped piloting and started shipping — and the gap is widening by the week.
OpenAI's latest usage data tells a story that should make every CTO sit up straight: agent usage in enterprise research workflows grew 56 times over in roughly seven months. Engineering use cases — the deep, mission-critical kind — grew 27x in the same window. These aren't vanity metrics from a handful of tech-forward unicorns. Enterprise clients now account for over 40% of OpenAI's total revenue, and that share is on pace to match consumer by the end of 2026.
Read that again. The same category that was being cautiously "explored" in boardroom slide decks two years ago is now approaching revenue parity with the consumer market that most people think of when they picture AI adoption. The structural shift isn't coming. It already happened.
What "Agentic AI" Actually Means in Practice
The term "agentic AI" gets thrown around a lot, so let's be concrete. Unlike a chatbot that answers a question and waits for the next one, an AI agent can plan a sequence of steps, use tools (databases, APIs, code interpreters, web search), delegate sub-tasks to other specialised agents, and loop back to check its own work — all without a human clicking "go" at each stage.
Think of it less like a smart autocomplete and more like a tireless digital colleague who never loses a file, never forgets a deadline, and — crucially — never complains about repetitive work.
In practice, this looks like:
- A financial services firm eliminating manual cross-platform data searches for 700 employees, saving 1,200 hours per month — hours that were immediately redirected to higher-value client-facing work.
- Engineering teams using multi-agent pipelines to autonomously handle code review, documentation generation, and regression testing across large codebases.
- Research and compliance functions running overnight analysis jobs that previously took analyst teams days, and waking up to structured, audit-ready reports.
These aren't science-fiction scenarios. They are live, in-production deployments generating measurable ROI right now.
The "We're Still Evaluating" Trap
Here's the uncomfortable truth: organisations still in the evaluation phase aren't being cautious — they're being lapped.
The competitors who moved early have already climbed the learning curve. They've hit the integration pain points, worked through the security questions, tuned their agent architectures, and are now on their second or third generation of internal deployment. By the time a late-mover finishes its proof-of-concept, early adopters will have compounded months of real-world iteration into a structural productivity advantage that is genuinely hard to close.
"The window for 'evaluating AI' has closed. The question is no longer whether to deploy agentic systems — it's how fast you can do it safely."
This is especially true in Switzerland, where industries like banking, insurance, pharmaceuticals, and precision manufacturing operate with stringent compliance requirements. The instinct to wait until everything is "perfectly clear" from a regulatory standpoint is understandable — but it's also a recipe for strategic irrelevance. The good news: compliance and speed are not mutually exclusive, provided you choose the right implementation partner.
Why Security and Compliance Can't Be an Afterthought
One of the most important lessons from early enterprise AI deployments is that bolting security on after the fact is expensive, slow, and often embarrassing. Agentic systems, by their nature, operate with broader access and more autonomy than traditional software. They interact with sensitive data stores, make decisions across systems, and sometimes act on behalf of users without a human in the loop.
In a Swiss context, this means considering:
- FINMA guidelines and data residency requirements for financial-sector deployments
- Swiss Data Protection Act (revDSG) obligations around automated decision-making and data minimisation
- Role-based access controls that ensure agents only touch the data they are authorised to touch
- Audit trails that satisfy internal governance and external regulatory review
Getting these layers right from day one is not a luxury — it's the difference between a deployment that scales confidently and one that gets quietly shut down after the first compliance audit.
The ROI Case Is No Longer Hypothetical
Let's talk numbers, because numbers have a way of cutting through strategic hesitation.
The financial-sector deployment mentioned above — 700 users, 1,200 hours saved per month — translates to something in the region of 14,400 hours per year returned to productive work. At even a conservative fully-loaded labour cost, that is a significant seven-figure annual saving. And that's a single use case in a single department.
Multiply that across research, operations, compliance, customer service, and engineering functions, and the picture becomes clear: agentic AI is not a cost centre — it's a force multiplier.
The 56x growth in enterprise agent usage didn't happen because organisations got swept up in hype. It happened because the early deployments worked, the ROI was measurable, and word travelled fast through procurement and strategy teams. That feedback loop is now self-reinforcing.
What "Moving Now" Actually Looks Like
For Swiss IT teams reading this and wondering where to start, here's a practical framework:
- Identify high-frequency, multi-step workflows — anywhere a human is manually stitching together information from more than two systems is a candidate for an agent.
- Quantify the current cost — hours spent, error rates, delay cycles. You need a baseline to measure against.
- Start with a bounded, high-value use case — not a sprawling transformation programme, but one workflow where success is clearly definable and measurable.
- Build security and compliance into the architecture from day one — not as a final review gate, but as a design constraint.
- Choose a partner who has done this before — real deployments, real integrations, real lessons learned. Not another slide deck.
The organisations winning right now are not necessarily the largest or the most technically sophisticated. They are the ones who decided — at a specific moment in time — that the cost of waiting exceeded the cost of moving. That moment, for most Swiss enterprises, is now.
The Infrastructure Decision You Can't Defer
There's a useful analogy in the history of cloud computing. For years, enterprises debated whether the cloud was "ready" for serious workloads. Then, quietly, the economics became undeniable, the tooling matured, and the conversation shifted from "should we?" to "how fast?" Organisations that made the call early built capabilities, reduced costs, and developed institutional knowledge that laggards spent years trying to replicate.
Agentic AI is at that inflection point today.
The data is in. The use cases are proven. The ROI is documented. What remains is an infrastructure decision — and infrastructure decisions, once deferred long enough, don't just delay progress. They define competitive position for years.
For Swiss businesses navigating this moment, the question isn't whether agentic AI belongs in your operations. It already does, whether you've deployed it or not — because your competitors have. The real question is how quickly you can close the gap, and who you want helping you do it.
The window is open. It won't stay that way forever.
Published in Stream · Dispatch #441 · July 2, 2026 · 7 min read.
Reply to paolo@mont3.ch - every email gets a human answer within 24h.