Part 1 of 3: Taste, Tools, and Infrastructure

In 2023, the bottleneck was production capacity. In 2026, the bottleneck is taste.

convergence

When AI handles creative work without human direction, everything starts to look the same. The tools optimise for what's most probable, and what's most probable is average, inoffensive slop.

The more people use the same tools in the same way, the more homogeneous the output becomes. You can see it across every creative field right now; there's a sameness creeping in that didn't exist three years ago.

the feeling

We used to spend the bulk of our time in creative work producing the thing. The strategic decisions (what to make, why, what's the point) took a fraction of the hours but accounted for most of the value.

AI inverts this. Someone with an idea and no technical background can now build and ship an app. The barrier that used to require a lot of resources (time and money) has collapsed.

The tools don't think about how things feel, or whether they're relevant. They handle the execution, but judgement just isn't part of their skillset. So when people use them without bringing that layer themselves, the output converges. All tools, no soul.

AI generates, it can't feel which one is right. But the feeling is the goal, right?

Taste is a practice. Designers have spent careers developing it; exposed to more input, context, cultural reference than most. They know what good looks and feels like, and how that evolves over time too. Which is why working with people who think this way, on the things that matter, still makes a real difference. They bring the taste layer that the tools can't.

three layers

I've started thinking about taste as three layers.

Direction. What is worth making? The brief before the brief. The "why this, why now, why us" that most teams skip entirely. Better decisions about what to execute.

Framing. What are the constraints? Constraints are creative fuel. Framing is the act of naming them and turning a feeling of friction into something you can build toward.

Curation. What makes the cut? Taste lives in the decision making.

So AI can handle the execution between the layers, but every layer requires human judgement. You stop being the person who makes things and become the person who decides what gets made.

why this matters

Most businesses are still running several disconnected AI tools with no shared memory between them. None of them know fully how the business works. OpenClaw, the autonomous agent that went viral this year, is already well past this, handling the bulk of work and sometimes even entire workflows for individuals and small teams. Big orgs haven't caught up.

I've been testing a different approach with clients - a coordination agent watches across everything the team uses, Notion, GitHub, wherever the work lives, and produces a morning brief: a plain-language update on what's happened since the last one. Right now it lands as a fresh page in Notion, a few checkboxes, and a text field for guidance on the narrative. The human directs it. The agent handles the rest.

What I'm building toward is a dashboard that makes this even more immediate, the intelligence already surfaced, waiting for a human decision. But even now, it replaces something that used to take weeks: getting a clear picture of what had happened, deciding what was worth saying publicly, briefing someone to write it.

the systems carry the weight

The bottleneck moved. It used to be: can we produce this fast enough? Now it's: does anyone here know what's worth producing?

The businesses that get this right are building systems that know the voice they use and the judgements they make, the things that take years to develop and can't be written in a manual. All built so the humans in the business can spend their time on the parts that require human judgement.

The systems carry the operational weight, so the people with taste can show up and use it.

A system with that depth of context doesn't just execute. It starts to notice patterns across the whole business that no single person ever could, because no single person touches all of it.

And if the person who built it leaves? The knowledge and the process stays.

If all this made you think of opportunities to upgrade your workflows, drop us an email to work with us.

This is Part 1 of a three-part series.

Part 2: Inside the Machine: What happens when you embed inside a non-technical business and build from observation.

Part 3: Tools vs Infrastructure: The business model that emerged, and why compounding systems are different from software.