If you're feeling uncertain about the future of AI agents, I don't blame you.
On one hand, OpenAI just unveiled agents so capable they'll apparently chase down an unpaid invoice before you've even noticed it's late. On the other hand, Gartner predicts that most enterprises will walk away from exactly this kind of AI setup within two years.
Same week, same technology, two completely different verdicts. So which one's right?
Read on and then leave your verdict in the comments below...
OpenAI Unveils "Dots," Always-On Agents That Work Independently
OpenAI has launched dots, a new category of AI agent built to run continuously in the background rather than wait for instructions. Powered by GPT-6 Astra, each dot gets its own cloud computer, learns a user's preferences over time, and connects to more than 4,000 apps through OpenAI's plugin ecosystem.
The idea is less "chatbot you prompt" and more "colleague who gets on with it."
As OpenAI CEO Sam Altman put it:
"It's like an AI helper that always has your back, inspired by the cool versions of what we all watched in movies growing up"
The examples OpenAI shared give a sense of that ambition.
-
A developer's dot watches customer feedback for recurring bugs, builds and tests the fixes, and hands over a finished pull request.
-
A sales lead's dot checks a client's requirements against product documentation, builds a proof of concept, and quietly updates the proposal as things change.
-
One early tester's dot even noticed he'd forgotten to invoice a client, prepared the invoice, and sent it once he approved.
Dots are rolling out now to Pro and Business Premium users, with specialist dots, versions carrying their own identity and system access, being piloted inside organizations and integrated with Microsoft Agent 365 for enterprise governance.
OpenAI's vision seems to be that the barrier to agentic AI was never capability; it was trust and access. Dots are its answer to both.
Gartner Predicts 70% of Enterprises Will Abandon Vendor-Built FDE Agentic AI by 2028
On the same day, Gartner published research that suggested a contrasting vision.
By 2028, the analyst firm predicts 70% of enterprises will abandon agentic AI built through vendor forward-deployed engineering (FDE), a model where vendor engineers build and deploy solutions directly inside a customer's business, because the costs become unsustainable and the enterprise never develops the ability to run the system itself.
Mukul Saha, Senior Director Analyst at Gartner, pointed to the real failure point.
"FDE engagements often fail structurally before they fail technically. The best-scoped FDE engagements have clear guidelines on governance, business value delivery, IP ownership, project co-ownership, knowledge transfer, and an exit strategy from day one"
Gartner's advice splits into three phases: scope the engagement properly before signing, embed vendor engineers with internal teams during delivery so knowledge actually transfers, and execute a genuine exit plan rather than extending the relationship because internal teams were never made ready.
The firm also flagged a more cynical risk, "FDE washing," in which ordinary consulting is rebranded as forward-deployed engineering to justify premium pricing without the delivery depth to back it up.
This pushes back against OpenAI's launch directly. Capability was never the real issue, it's whether a business can actually own and build on what it buys. That's where Gartner thinks most companies get it wrong.
My take? The future will see 'citizen developers' in every enterprise team. This will naturally reduce overreliance on vendors. But will it see 70% of enterprises abandon vendor FDE? Within two years? That might be a stretch.
In other news: Accenture's Earnings Reveal Just How Much AI Is Actually Driving Enterprise Spend
Accenture closed fiscal 2026 with revenue up 6% to $74.2 billion, and AI is clearly doing the heavy lifting. Nearly 100 clients started their first advanced AI engagement in Q4 alone, pushing the full-year total past 400, while bookings from its AI and data partners more than tripled. CEO Julie Sweet credited the growth to clients rebuilding the "enterprise AI stack they need to use AI at scale," backed by a workforce that just hit 110,000 AI and data professionals, a year ahead of schedule.
The result also lands squarely in this week's FDE debate. Asked directly about Palantir's forward-deployed engineering model, Sweet called FDE "a growth opportunity," something Accenture simply scales as a repeatable offering. That's a notably different posture to Gartner's warning that vendor-built FDE needs a built-in exit plan, Accenture is betting it can absorb that complexity permanently rather than hand it back.
Accenture isn't just riding the AI wave, it's betting on staying inside the client's walls indefinitely. If Gartner's prediction holds and enterprises start pulling back from that kind of dependency, Accenture's growth numbers today could look very different in a couple of years.
Barclays Shows What Disciplined AI Scaling Actually Looks Like
Barclays provides an interesting test case because it sits between OpenAI's vision of widespread agentic AI adoption and Gartner's concerns around ownership and governance. Rather than outsourcing capability, the bank appears focused on embedding AI inside existing teams and processes.
Barclays is expanding its use of Anthropic's Claude across its global operations, with Claude Code adoption expected to reach 50% of its developer population by the end of 2026, and a majority by 2027.
The bank's Colleague Knowledge Assistant, live since 2025 and built on Claude through a retrieval-augmented generation setup, has been adopted by more than 16,000 colleagues and handled over a million searches, helping staff find answers faster for Barclays' 20 million UK retail customers. Separately, Claude models now classify and route roughly 120,000 incoming client emails a day within the bank's Global Markets business.
Anne Marie Darling, Group Co-Chief Operating Officer at Barclays, highlighted the governance angle:
"Expanding the use of Claude across Barclays is about more than adopting AI. It is about transforming how work gets done... within a secure and well-governed environment"
Anthropic's Chief Commercial Officer, Paul Smith, shared:
"Claude now helps 16,000 Barclays' colleagues find answers for customers, sorts 120,000 emails a day, and will be in the hands of most Barclays engineers by 2027. Very few institutions hold AI to such a high standard."
In Brief
Gather Launches AI Customer Simulations for Faster Insights
Gather has launched customer simulations built from AI-moderated interviews with real people, giving teams evidence-backed insight. The platform continuously fills research gaps, helping CX, marketing, and product teams test messaging, understand customer needs, and make faster decisions without relying on traditional research processes.
NVIDIA Launches Open Agent Safety Platform for AI Agent Governance
NVIDIA has launched the Open Agent Safety Platform, combining OpenShell runtime software with Sentry, a hardware-based monitoring design. The system is meant to set enforceable boundaries outside the model itself, with OpenShell tracing actions and Sentry quarantining agents that breach limits within milliseconds.
In Case You Missed Last Week:
Voice AI Lawsuits Put Contact Centers on BIPA Watch
A fresh warning from the Contact Center AI Association points to a sharper legal risk for voice AI, with new claims against Walmart and Lowe's pushing BIPA litigation closer to everyday customer service. Rob Wilkinson breaks down when call recording tips into biometric processing, and what contact center buyers need to check before they get caught out.
How To Manage an Ecosystem of AI Agents in Your CX Stack
As businesses stack up customer-facing bots, internal copilots, and operations agents, Fin's Brian Donohue tells CX Today how to stop that ecosystem from recreating the data silos AI was meant to fix, and where customer trust in AI still falls short.
