Morse genuinely contributed a smart component to the telegraph system, the relay, that made it work long distance. A man called Harold Arnold, picking up where Lee de Forest left off, improved the audion, transforming it into the hard valve or vacuum tube. Without the tube, not plugged in until well after Alexander Graham Bell's patent expired, the telephone could not work long distance and thus could not compete with, let alone replace, the telegraph. I do wonder if something has just happened at Anthropic.

A few weeks ago, Chris and I, as we recorded our first episode of Container Solutions TV, spoke about Claude's structured outputs. Will they change the game like the relay and tube once did? That's what I have been thinking about ever since. Structured outputs, Chris explained to me, allows programmers, through JSON schemas, to interact with LLMs not like they were chatbots but as if they were deterministic, like a Java library might be.

'OK', you might be thinking, 'why does that matter'?

Until recently, LLMs have mostly been human facing, or as Chris put it on the show: we're not talking about a user-to-computer interface but a computer-to-computer interface. Structured outputs mean an LLM can now return validated data, such as first name, surname, DOB, etc., in a way the called software expects and developers can thus build software against that expectation. This is, incidentally, what makes agentic AI possible, because agents don't just chat but act by calling APIs, reading databases and triggering workflows. Agents and agentic AI won't work without LLMs that behave like deterministic software components. Structured outputs are what will make that happen.

Which brings us back to a favourite theme of ours… Throughout Visionaries, Rebels and Machines is the idea of the third inventor. The third inventor, if you haven't read the book, maps a new technology to an often unarticulated and unknown user need. It is in many ways the final stage of innovation and it requires creativity that is different to the creativity needed for rearranging components into a system, which is in turn different from the creativity involved in the continuous improvements needed to bring a component to life. Edison's grid had a clean enough interface, like Anthropic's structured output, which gave third inventors the chance to invent applications, which ranged from toasters to fridges, that plugged into it. Bell Labs gave samples of the transistor out knowing others, the third inventors, would work out applications they could not, since they thought the transistor would only find applications in a hearing aid. LOL.

The third inventor leads to the theme of the book, namely that:

The destiny of a general-purpose technology lies not in the hands of its creator but its users, who seek out applications that were previously unimaginable.

Edison, like the boffins at Bell Labs, had no idea what users would do with his inventions in the same way that Berners-Lee had no idea what Bezos and Hastings would do with his. We have no idea what creative users, like the team at Container Solutions, will do with Claude's structured outputs but I guarantee, with history on my side, they will do something with them.

Kierkegaard said life has to be lived forward but understood backward. The same is true for technologies, we invent them before we know what to do with them.

Chris mentioned it on the podcast as gossip, understated, boring even. But this is the moment the third inventor's race began. Buckle up, buttercup. The game is afoot.

Further Reading

The podcast is out now for those who want to listen along in the gym or whilst walking the dog. Check out Apple or Spotify or your favourite platform.

Morse's relay story can be found in How Innovation Really Works.

Finally, here is Chris over on YouTube if you want some lunchtime viewing: