A decade ago, we mapped the landscape of 2-speed IT and identified a critical problem: the trough of disillusionment between what businesses needed and what IT could deliver. The gap between Digital Speed and Operational Speed was widening, and “acceptable disappointment” had become unacceptable.
Fast forward to late 2025, and I’m watching history repeat itself. There’s a new trough forming around agentic AI. The hype cycle has peaked, reality has set in, and the whispers have started: “It doesn’t work as promised.” “Too unpredictable.” “Can’t rely on it for production.”
Here’s what concerns me: we’re misdiagnosing the problem. Again.
The Same Mistake, Different Technology
The original 2-speed problem wasn’t that digital transformation was impossible—it was that IT didn’t know how to deliver it reliably. The patterns, practices, and mindsets that worked for Operational Speed failed spectacularly for Digital Speed. It took years for the industry to discover that agile wasn’t just faster waterfall, that DevOps wasn’t just automation, that cloud wasn’t just someone else’s servers.
We’re making exactly the same mistake with agentic AI.
The disillusionment isn’t because agents, or multi-agent teams, don’t work. It’s because we’re trying to deploy them using the same delivery models, the same governance frameworks, and the same organizational structures that we’ve always used. We’re surprised when they fail.
Software Development, As We Know It, Is Dead
Let me be direct: software development as we’ve practiced it for decades is dead. Not evolving. Not improving. Dead.
For twenty years, the model was consistent: hire developers, organize them into teams, manage their output through backlogs and sprints. Whether onshore or offshore, human capacity was the constraint, and labor cost was the lever.
Agentic AI shatters that entire framework. When properly deployed, human capacity isn’t a constraint anymore. But—and this is crucial—only if you’ve discovered the patterns that work.
Most organizations are bolting AI agents onto their existing delivery model: same team structures, same backlog management, same handoffs and status meetings, and quarterly planning cycles. They’ve added AI agents as if they’re just another resource to staff onto a Jira board.
It doesn’t work. It can’t work. The entire operating model needs reimagining.
Pattern Recognition: What Actually Works
Through building our platform and working with early adopters, we’ve discovered patterns that consistently deliver results. AI in software development is the bellwether—the most visible indicator of what changes—but the implications ripple through the entire operating model.
Yes, agents can accelerate existing processes—having digital workers handle repetitive tasks faster and more reliably. That’s valuable. That’s also table stakes. It’s the difference between using the cloud to host your old applications versus rearchitecting for cloud native. Both work, but only one fundamentally transforms the economics and capabilities of what’s possible.
Work gets reorganized around outcomes, not tasks. In Agentic AI powered software development, agents don’t replace developers on teams—they replace entire delivery pipelines. Look broader: the same applies everywhere. Customer service, financial operations, and compliance monitoring. The unit of work shifts from “who does this task” to “what outcome needs achieving.”
What Agentic AI Powered Software Development Actually Changes
Human roles fundamentally shift. Developers aren’t writing less code because agents help them code faster—they’ve stopped writing implementation code. They’re defining outcomes, establishing guardrails, and curating what agents produce. Apply that lens everywhere. Finance teams aren’t processing invoices faster; they’re not processing them at all. They’re governing exception handling. HR isn’t screening more candidates efficiently; they’re orchestrating how talent matching happens.
Speed becomes the baseline, not the aspiration. In 2014, “days and weeks not months and years” seemed revolutionary. With properly deployed agents, if you’re measuring operational delivery in days, you’re behind. And this isn’t just software releases. It’s financial closes, compliance reporting, customer issue resolution, and strategic analysis. The tempo of the entire operating model accelerates.
Quality comes from architecture, not review. Traditional quality control becomes a bottleneck when agents produce implementations in minutes. Quality must be embedded in how agents are constrained and directed. You can’t have humans reviewing every agent’s action, every decision, or every output. Your operating model must embed quality in the orchestration, not bolt it on through oversight.
The economics of scale invert. Small teams with agents deliver large traditional teams. But this isn’t about the development of productivity; it’s about organizational, functionality, and productivity. The calculus around team size, span of control, organizational layers—it all changes.
Why Organizations Haven’t Grasped Agentic AI Business Transformation
Most organizations are treating agentic AI like any other technology adoption: pilot projects, governance committees, risk assessments, phased rollouts. They want to measure agent productivity the way they’ve always measured productivity. They want to manage agent work the way they’ve always managed work.
This is a category error. Like measuring Digital Speed using Operational Speed metrics.
But here’s the deeper issue: organizations are approaching this as a technology question when it’s a business operating model question. Software development is just the most visible manifestation because the transformation is obvious and measurable. The real opportunity is to recognize that if the way we deliver software has fundamentally changed, then the way we operate the entire business can fundamentally change.
The organizations succeeding are those willing to start from first principles. They’re not asking, “how do we add AI agents to our current model?” but “if we could organize work any way we wanted, what would we build?”. What would true Agentic AI Business Transformation look like?
The Platform Imperative
You can’t just give teams access to Claude or ChatGPT and expect transformation. That’s like handing someone AWS credentials and expecting them to build modern cloud architecture.
What’s needed is a platform that embeds the patterns that work:
- Multi-agent orchestration that coordinates specialized agents toward outcomes
- Guardrails and constraints that ensure quality without manual review
- Integration with existing systems without wholesale replacement
- Observability that shows what agents are doing and why
- Governance that works at machine speed, not meeting speed.
This is what we have been building – a business platform that utilizes multi-agent teams to do work, but in a way that reimagines what work needs for the future.
The AI models themselves are a commodity infrastructure now. What matters are the patterns, practices, and platforms that enable reliable execution across the operating model.
We started with AI powered software development because that’s where the patterns are most visible and benefit most immediately. But the platform we’ve built doesn’t care what domain it’s applied to. The same orchestration that coordinates agents to deliver code can coordinate agents to close financial periods, resolve customer issues, or monitor compliance requirements.
Don’t Fall Into the Trough of Disillusionment
The risk isn’t that agentic AI fails to deliver. It’s that organizations give up before discovering how to operate with it reliably.
In 2014, many companies concluded that “digital” wasn’t for them, that their industry was different, that the old ways were good enough. Competitors who figured it out disrupted them.
The same will happen here. The trough of disillusionment around agentic AI is real, but it’s temporary and survivable—if you recognize what’s actually happening.
The gap between business need and operational delivery hasn’t widened because of AI. The gap can finally close, but most effectively for organizations willing to fundamentally rethink their operating model, not just their technology stack.
The patterns exist. The platforms are emerging. The organizations that dismiss agentic AI as overhyped technology will watch from the trough as competitors transform their economics, their speed, and their entire operating model.
We spent a decade bridging the gap between Operational and Digital Speed. Don’t spend another decade making the same mistakes with AI.
The trough is there. But now we know how to cross it, through Agentic AI powered business transformation in the Machine Speed era of Business.
Reflecting on lessons from 2-speed IT and observations on how the industry is approaching—and mostly missing—the transformation potential of agentic AI in 2026.
References:
- 2-Speed IT was a concept described in a published white paper by Greg Smith and Rob Price whilst they were at Atos Consulting. Rob extended it to 3-speed, in a later article published here, whilst he was COO at Worldline UK&I & CEE.
Note: This article was written with the assistance of Claude, in an iterative conversation with Rob Price. It was written for publication in the US and the UK markets, and the choice has been made to write in US English.
