Part 1 of 3: Agents are eating the organisation

In 2011, the venture capitalist Marc Andreessen made a prediction that turned out to be prophetic in almost every respect, except one. Software did eat the world but it did not eat the organisation. That omission is what this series is about.

The argument Marc Andreessen made in August 2011, in a Wall Street Journal essay, that arguably became the defining document of a generation of technology investment, was essentially this: the world was undergoing a broad and irreversible shift, one in which software-native companies would restructure industry after industry, not by competing on the margins of existing markets, but by replacing their operating logic entirely. To that extent, he was correct. Retail, music, television, transport, accommodation, each was transformed, and faster than most analysts anticipated.

What is worth reflecting on however, fifteen years later, is precisely what software did and crucially did not eat. Andreessen’s revolution was fundamentally about distribution, that is the interface between organisations and the world around them. Amazon used software to restructure the relationship between buyers and sellers. Spotify used it to restructure the relationship between listeners and music. Uber used it to restructure the relationship between passengers and drivers. In each case, the transformation happened at the boundary, between the organisation and its market, its customers, its supply chain.

Inside the organisation, in the structure of teams, the logic of processes, the design of how decisions are made and work is coordinated, the story is considerably less dramatic. Look at how most organisations actually function today. Many Finance teams still run a month-end close on the same cadence and cognitive architecture they used in 1995, with Excel where there was once paper. Many Sales teams still follow relationship sequences that a 1970s account manager would recognise. The software arrived at the boundary. The interior of the organisation, for the most part, did not change.

This is not a failure of adoption. After all, organisations have invested heavily in enterprise software (CRMs, ERPs, HRIS platforms, finance systems) for decades. The investment is real and so are many of the benefits. But there is a difference between deploying software within an existing organisational logic and redesigning that logic around what software makes possible. Almost everywhere you look, organisations have done the former. Almost nowhere have they seriously attempted the latter.

The Constraint

Conway’s Law and the Shape of Organisations

In 1967, the American computer scientist Melvin Conway observed ‘any organisation that designs a system will inevitably produce a design that mirrors the communication structure of that organisation’. He was writing about software development, but the principle generalises. The organisation is not neutral terrain on which technology lands. It is the mould. And if the mould does not change, neither does what comes out of it.

Conway’s Law has a corollary that matters for today. If the organisation shapes what technology produces, then deploying more powerful technology into an unreformed organisational structure does not unlock its potential, it in fact amplifies the structure’s existing characteristics, including its dysfunctions. A finance function designed around monthly human review cycles will produce monthly human-paced outputs, regardless of how much AI is embedded in the process. Where technology accelerates, the organisational logic itself can constrain.

This is the gap that the first wave of software left open and it is the gap that matters most for what is coming next. Because unlike the software of Andreessen’s 2011 essay, which sat primarily at the boundary between organisations and the outside world, the next generation of AI does not operate at the boundary. It operates inside — in the middle of processes, embedded in the logic of how work is actually structured and decisions are actually made. For the first time, the technology has reached the place where Conway’s Law is most active.

Which means, for the first time, that changing the technology without changing the organisation is not just a missed opportunity, it is the guaranteed route to reproducing existing limitations at scale and speed.

The Measure

What the Productivity Statistics Actually Show

Robert Solow’s oft quoted remark — “you can see the computer age everywhere except in the productivity statistics” which seemed wry at the time has turned out to be the beginning of a thirty-year research programme. Computing had been embedded in Western economies since the 1960s. Mainframes, minicomputers, personal computers, wave after wave of genuine capability, accompanied by substantial investment and yet measured productivity growth, by historical standards, was flat.

The concept Solow formalised in 1956 is the right frame for understanding this. Total Factor Productivity (TFP) is what remains when you subtract the contributions of labour and capital from economic output. It is the measure of how cleverly an economy uses what it has, the residual that captures better processes, better organisation, better deployment of knowledge. For decades, TFP growth had been disappointing almost exactly where digital investment had been most enthusiastic.

The UK’s record makes the point with contemporary force. Before the financial crisis, output per hour was growing at over 2% annually. Since 2008, the average has been closer to 0.5%. The Office for National Statistics estimates that if pre-crisis trends had continued, output per hour would now be more than a third higher than it is. That is not a statistical footnote. It is the accumulated cost of an economy that has adopted substantial technology without generating commensurate productivity gains from it.

The technology, clearly, has not been the problem. The problem is something more structural, something that connects directly to the gap Conway’s Law describes. And understanding it requires looking not at the present, but at history.

1
At Joulen, our response to Conway’s Law has been practical as well as theoretical. Our engineering teams are structured around the Team Topologies framework — an organisational design model that deliberately aligns team boundaries with the architecture of the systems being built, rather than allowing organisational structure to impose itself on technical output by default. Thanks to our Head of Engineering, Andy Brownlie, for leading that thinking.

In the next post, we examine why this pattern, technology arriving decades before its productivity dividend, is not new. It has happened before, with steam, with electricity, and with computing. The lessons from those transitions are the most useful guide we have to what is happening now.

Exhibit 1

UK multi-factor productivity growth, market sector — annual % change, 1998–2024. The collapse after 2008 has never fully recovered, despite continuous technology investment.

Source: Office for National Statistics, Annual Multi-Factor Productivity, Market Sector, May 2025