https://gwern.net/doc/economics/automation/1990-david.pdf

Realistically, AI is only going to show productivity gains for the approximately 2% involved in R&D. These jobs involve generation of new ideas, processes, and products and can make use of the force amplifier. But unless most orgs grow R&D from 2.2% revenue to say 5%, we won't see the gains in real terms. My personal theory is the AI TAM is already flooded, having consulted in this space for the last 8 years, and worked in product R&D for 26.

@zarazhangrui cost of overhaul > cost of integration > cost of status quo. Until the signs flip you won't see this expectant change. Same reason why robotics is now finally being integrated. increase minimum wage and it finally made robotics financially make sense.

@zarazhangrui During the Business Process Re-Engineering era, this was known as "paving cow paths"

Bain recently reported on the lackluster savings from early AI implementations. Their top observation: "The single most costly mistake in AI deployment is automating a broken process."

@zarazhangrui @v_vashishta Where is the news here? A 200 year old mistake is getting repeated?

New AI in old processes adds complexity and thus costs.

The same is true for humanoids.

Unless you build a new AI-native or AI-enabled process you fail your business.

@zarazhangrui David's real point is that the gains came from re-laying out the whole factory around unit drives, not the motor swap. The AI equivalent is rebuilding workflows around agents, which is exactly the work most orgs skip. Bolt a copilot onto a 2019 process and you get nothing.

@zarazhangrui The Dynamo analogy cuts deep. Companies bolt AI onto existing workflows instead of rearchitecting around autonomous execution. Agent infrastructure (like what we're building @clark__labs) is the electrification of the process itself.

@zarazhangrui I'm seeing this more and more in AI deployments.

I used to think the Central Brain was a nice to have, but it really needs to be structurally in place from the start.

Any other foundation would be localized optimization at best.

@zarazhangrui Agree and see my writing recently:

The dynamo lag had a tangible signal — you could physically see whether a factory had moved from group drive to unit drive. AI has no visible equivalent. The redesign happens in role definitions, decision rights, and process documentation. All invisible to leadership. Which is why most companies will keep declaring "transformation done" while running the same workflow with a faster steam engine.

@zarazhangrui This is a good analogy, but the problem still remains. Large incumbents are always locked in to older systems, workflows and technologies. Changing fundamentals is very very dangerous to business (1)

The key insight from Paul David: gains only came when factories were redesigned around the motor, not when they swapped the steam engine for an electric one. Same bet for AI — companies that redesign workflows from scratch around LLMs will look like a different species from those that just plug in a chat widget. The reorganization lag is real, and probably longer than most expect.

@zarazhangrui The steam engine was a tool. The factory redesign was the transformation. AI is the same, the real upside comes from changing the workflow, not just speeding up the old one.

@zarazhangrui @drishtadyumn It took a depression to find those techniques just saying

@zarazhangrui Great parallel. Healthcare is stuck in exactly that phase - just swapping the steam engine for electric. When do we get to redesign clinical practice around what AI actually enables? 🧬

@zarazhangrui each machine got its own small motor.——each phone gots its own open source world model!

@zarazhangrui @mjasay Same thing with web pages. Early versions were copies of print. Took several years for the designs to adapt to the new interactive capabilities enabled by the web.

@zarazhangrui that why companies building ai integrated with their product from scratch will win traditional company trying to just add a chat interface sidebar

@zarazhangrui Thats why young entrepreneurs and new companies have an advantage. You have the chance to design the factory the right way, rather than trying to salvage a mammoth