When computing lived in institutions, not pockets
In the 1960s, a computer was infrastructure, not a consumer good. The Apollo Guidance Computer, built by MIT for NASA, was groundbreaking precisely because it squeezed navigation, engine control, and error recovery into hardware small enough to fly — a "considerable advance in miniaturisation" for its time. It also pioneered ideas that still underpin modern software: task prioritisation, real-time error handling, and the kind of resilient design that let the system recover from an overloaded processor during the actual lunar descent, rather than crash. But none of that capability existed in isolation. The computer worked because it was embedded in a much larger system of trained astronauts, ground engineers, and a mission control room feeding it data and judgement in real time — a reminder that even the most celebrated "computing breakthrough" of the era was, in practice, a human-machine partnership. That framing matters, because the same will be true of AI: the technology’s usefulness is never separable from the people and processes around it.
For most of the following decades, computing stayed largely institutional — mainframes in banks, systems in corporate back offices, the slow build-out of internet infrastructure. It was powerful, but it was not personal. Computer Weekly’s own retrospective on its 60 years traces this arc: technology moving "from institutional breakthroughs to personal technology," step by step, as it embedded first in enterprise systems and financial markets, then in corporate infrastructure, then through the internet.
The smartphone: computing becomes a habit, not an event
The real turning point for everyday life came with the smartphone. As Computer Weekly’s anniversary essay puts it, the 2007 launch of the iPhone "changed the centre of gravity" — computing moved out of specialist environments and into people’s hands, reshaping how people communicate, shop, and consume information. This wasn’t just a hardware milestone; it created a new economic layer. Mobile connectivity and app ecosystems turned computing into something ambient — always on, always nearby — which in turn made it possible to build entirely new business models around subscriptions, on-demand services, and platforms that took a cut of every transaction they mediated. The global financial crisis that followed accelerated this shift further, pushing a new generation of founders to build differently as older industries contracted. What mattered commercially wasn’t simply that phones were powerful computers — it was that they were personal, continuous, and integrated into daily habits in a way institutional computing never was.
That is the pattern worth holding onto as AI moves through its own version of this transition.
flowchart TD A[Apollo era: mission-critical computing, 1960s-70s] --> B[Smartphone era: personal, always-on computing, 2007 onward] B --> C[AI era: computing embedded in everyday workflows, today]
What changes when a technology becomes mainstream
Each of these stages didn’t just add processing power — it changed who used the technology, where it lived, and what actually limited its value. Laying the three stages side by side makes the pattern clearer:
| Dimension | Apollo-era computing | Smartphone era | AI adoption today |
|---|---|---|---|
| Who used it | Government agencies, research labs, engineers | General consumers, workers, small businesses | Employees, teams, entire organizations |
| Where it lived | Dedicated hardware for a single mission | A personal device carried everywhere | Embedded inside existing software and workflows |
| What it primarily solved | Navigation, control, and coordination for a singular engineering feat | Communication, information access, on-demand services | Search, summarisation, drafting, and, in some cases, decision support |
| Main constraint on value | Scarce hardware, cost, and specialist expertise | Network coverage, device affordability, app ecosystem maturity | Organizational readiness: skills, training, trust, and process redesign |
The common thread is that raw capability was rarely the bottleneck for long. Once the underlying technology worked well enough, the harder — and more commercially decisive — problem became integration: fitting the tool into how people already worked and lived, and building the surrounding structures (finance, skills, regulation, culture) that let it scale responsibly.
AI’s second migration: from experiment to workflow
Computer Weekly’s essay argues that AI is "now being deployed across almost every sector of the economy, from financial services and healthcare to logistics and creative industries," and that "the focus has moved from experimentation to implementation". That’s a meaningful claim, but it’s worth being precise about what it does and doesn’t mean. Deployment is not the same as durable value. Plenty of organizations have rolled AI tools into daily use without yet reorganising how work actually gets done around them — which is exactly the gap independent workforce research has started to quantify.
A large 2025 global survey of employees and employers found that 88% of workers already use AI at work, but mostly for basic tasks like search and document summarisation; only 5% use it in ways that meaningfully transform how they work. The same research suggests that companies may be leaving as much as 40% of potential productivity gains on the table because of gaps in training, culture, and workforce readiness rather than gaps in the technology itself. Tellingly, only 12% of employees report receiving enough AI training to unlock its fuller benefits, even as 64% say their workloads have grown under the pressure to use these tools. This is a strikingly familiar pattern: a powerful capability arrives faster than the organizational habits needed to use it well — much as smartphones outpaced many businesses’ ability to redesign services around mobile-first customers, and much as the Apollo computer’s raw power meant little without trained crews and a coordinated mission around it.
The bottleneck is organizational, not technical
None of this means AI’s trajectory is guaranteed to repeat the smartphone’s runaway diffusion, nor that regions currently seen as AI hubs — London among them, given its concentration of finance, research, and engineering talent — will necessarily hold that position indefinitely. Ecosystem strength is not the same as permanent advantage, and enthusiasm about deployment numbers should not be mistaken for evidence that value is already being captured at scale across most sectors.
What the historical arc does suggest is where to look for real progress: not in headline model capability, but in whether organizations build the skills, training, governance, and inclusive design practices that let a general-purpose tool actually change how work gets done. Computer Weekly’s essay makes a similar point about the human side of the equation, noting that skills development and diverse, representative teams will shape how well AI’s influence is directed.
The pattern, not the hype
The story of computing since the 1960s is not one of a single dramatic leap but of repeated, uneven migrations — from institutions to pockets, and now from pockets into the fabric of daily work. Each stage looked, in hindsight, inevitable; each was in fact contested, unevenly adopted, and dependent on unglamorous groundwork — training, trust, infrastructure, coordination. AI is currently somewhere in the middle of its own migration: widely used, unevenly mastered. Whether it ends up reshaping work the way the smartphone reshaped consumer life will depend less on how clever the models become, and more on whether the people and organizations around them are ready to change how they operate.
