• School of Cybernetics; livecoding; fifteen years teaching programming, now teaching agentic coding (students + LLMs Unplugged this arvo)
  • my angle: not what the models can do, what changes about the work
  • pick up from Lexing’s adoption curve, zoom in on one segment
  • the IR deal: it made pretty good cheap. Tool up once, stamp out thousands. The price was fit: the factory’s sizes, not the tailor’s cut
  • software ran the same deal, harder: the first copy is the whole cost. So SaaS: sell it ten thousand times, everyone bends their process to fit
  • this room’s software estate is off the rack all the way down. It was the only deal on offer
  • agentic coding collapses the cost of the first copy: idea to working thing in an afternoon
  • bespoke at the off-the-shelf price. This revolution runs the IR deal the other way: exactly right, cheap
  • precise: the first working version gets cheap, not “software”
  • but a tailor needs a tape measure: calendars, mail, files are where this place’s real processes live
  • concede it: ten thousand hand-made tools is a real maintenance problem, and caution about agents near our data is fair
  • but mind the gap between “embrace AI” and what lands on the desktop. We were promised a tailor and mostly got a personal shopper
  • helpful, friendly, knows the stock — and it can only take you back to the rack. No agentic tools, no access to the data, nothing it can measure or cut
  • a personal shopper helps you choose what fits best. A tailor makes it fit
  • all the enthusiasm, none of the agency
  • the opportunity if we hand over the tape measure: more software, made by people who know how their unit works. Home-cooked software; a chance to make good software again
  • making was the bottleneck; now the constraint is knowing what you want, and saying why. Those people are in this room, not the IT shop
  • govern with judgement, not prohibition. It’s a crit: pin it up, say why this one fits
  • a choice you can’t argue for is a default — and that cuts both ways, for what we switch on and what we switch off
  • closer: if you think AI is going to change the world, but you can’t imagine it changing your default office productivity suite, you’ve got a much narrower idea of “changing the world” than you think
  • a personal shopper helps you find your way around that suite; a tailor changes it
  • (backup if needed: Satya Nadella, BG2 podcast, Dec 2024 — business apps “will all collapse in the agent era”, and “why do I need Excel?” His answer: the agent drives Excel for you. Only works if the agent is allowed to touch it)