Previously at the Volcano Base I’d been observing how AI shifts from being an assistant to a representative when it’s given a body. Since then, I was asked to be guest speaker at a SaaS company conference. The gist of that talk follows below.

Most conversations about AI start in the wrong place.

They begin with speed. Productivity. Output per head. The hope that everyone can suddenly do more work in less time, preferably without changing anything else.

That framing sounds sensible. It’s also misleading.

AI doesn’t multiply effort. It multiplies the distance intent can travel before it dissipates.

At this point, most people are already using AI in some form. Drafting. Summarising. Asking questions they don’t want to ask a search engine.

That’s adoption.

What hasn’t changed, in most organisations, is how work actually moves. The handovers. The waiting. The moments where context has to be reconstructed because it fell out of the system somewhere between tools, people, or teams.

If AI sits outside the work, it stays optional. And optional things rarely change outcomes.

Work rarely fails inside a role. It fails between them.

Between people. Between tools. Between “what I meant” and “what you received”.

This is where context leaks. Where data gets re-entered. Where decisions get revisited because the reasoning disappeared into a document, a message, or a spreadsheet that exists purely to hold the why together.

Org charts don’t show this. Process maps smooth it out. But everyone recognises it instantly in their own work.

Friction lives in translation and coordination.

When systems don’t quite work, people adapt.

They build workarounds. They learn who to message. They remember the unofficial steps. Over time, we stop calling this compensation and start calling it experience.

It’s valuable. It’s respected. And it’s doing a job systems never quite managed to do.

For a long time, humans were the only thing that could hold intent together across broken flows. We quietly used human cognition as glue.

The real shift with modern AI isn’t better answers. It’s memory.

Systems can now hold more context, for longer, across multiple steps of work. When context stops disappearing, decisions change. Not just how fast they’re made, but what “a good decision” even means.

At that point, AI stops being something you dip into. It starts becoming a persistent part of the system itself.

This is where the phrase “force multiplier” tends to mislead. We imagine people working faster. That’s not where the big gains come from. Most productivity gains come from subtraction.

Less chasing. Less translating. Less resetting context. Fewer loops where intent quietly evaporates and has to be rebuilt by hand. Remove those points of friction and speed appears as a side effect.

Most organisations don’t lack intent. They lack systems that can carry it intact.

So the interesting question isn’t how much effort AI can save. It’s where, in your own work, intent tends to disappear.

Some interesting stuff I discovered on my adventures…

Anthropic’s Claude Cowork research preview turns an assistant into something that can work inside tasks and files rather than waiting to be prompted. Why it matters: this is AI reducing friction at handover points by carrying context forward instead of forcing constant restarts.

OpenAI’s ChatGPT Health is a specialised version designed to help people navigate medical information with tighter safeguards and clearer boundaries. Why it matters: a glimpse of how intent can be supported without pretending the system replaces judgement or responsibility.

Apple says its next-generation AI-powered Siri will rely on Google’s Gemini language models rather than something built entirely in-house. Why it matters: even Apple is choosing leverage over ownership, a reminder that systems carry intent only as well as their architecture allows.