I was wrong about my own idea, and I found out in an afternoon instead of a quarter.
I’m currently heads-down building a Startup in the fintech space and last week I had an idea for a new piece of functionality. I thought I knew exactly what it should do and how it should do it. It ends up that I was wrong. Very wrong.
Not wrong in the sense of sloppy. Wrong in the sense that I had never used the thing so there were unknowns that I hadn’t considered. My initial thoughts described an idea that had never seen the light of day.
But because I’m now able to build instead of describe, after creating a prototype I knew a LOT more about what I wanted. Not because I thought harder about it, but because I built it, clicked it, hated it, and changed it over and over again. Then I shipped it to the team to get their reaction. The conversations we had were about a working feature instead of a paragraph describing one.
Everything we call communication at work is Proxy Work
Proxy Work is the artifact you make instead of the thing. The wireframe instead of the page. The spec instead of the feature. The deck instead of the decision. The model instead of the business.
We’ve been doing it so long that we think of it as what we get paid to do.
Go back to General Motors in the 1930s. Harley Earl started sculpting full-size cars out of clay (actually plasticine), and almost a century later nearly every car studio in the world still does it. It’s a crazy process that you should watch a video about if you haven’t seen it. It requires a warm room to keep the material soft, a giant five-axis mill to sculpt with, then countless hours of hand finishing. It adds up to weeks of work on a car with no engine, no seats, and no chance of ever moving under its own power.
Why go to the trouble? Because cutting steel tooling costs millions of dollars, and you get one shot at it. The clay exists so that fifteen people can stand around it and argue before anyone spends the money.
But nobody ever gets to drive the clay.
This is the deal every knowledge worker has implicitly accepted their entire career. The real thing is too expensive to use as a conversation starter, so we build a cheap fake, argue about the fake, and hope the argument transfers. Whiteboards, PRDs, Figma files, PowerPoint, and the spreadsheet with the assumptions tab. Everything is clay.
The proxy was ALWAYS worse than the thing. And people’s effort and contribution to the proxy work of a project is ultimately thrown away. We use it when building the thing is expensive. That’s the only good reason.
Ten iterations can buy a different kind of confidence
I now communicate in code, which sounds more clever than it is. Nobody on my team reads the code and I certainly don’t read the code. Code is the substrate, the way ink is the substrate for a physical book. What lands in my team’s hands is a working mobile app, or a marketing campaign that feels real, or a piece of functionality they can step through to see how it feels through the lens of a user. The thing itself has become cheap enough to send as a first draft, which is a more profound idea than “AI makes you faster.”
In my opening example, if I had shipped a spec to the team, we would have spent a week debating the details of a version of the feature that nobody would have wanted once it existed. The team would have built it, we would have discovered it only kind-of-sort-of worked, and then we would have specced out a set of fixes, and then the cycle would start over again. And again. And again.
Iteration used to be expensive, so we rationed it. The whole apparatus of product management exists to squeeze as much certainty as possible out of a small number of expensive guesses. When iteration is cheap, you should stop guessing. You can play with the thing until you know, and then you’ll be able to communicate from a position of actually knowing, which is a different message than the confident one you send when you are describing something that looks amazing in your head. The irony of that sentence is not lost on me because I have no mind’s eye (I have a condition called Aphantasia). I cannot picture anything in my head at all, which may explain why I’ve been leaning hard into the concept of building first and nailing down details later.
Working is not the same as decided
There is a pitfall in all of this, and a lot of teams are going to walk straight into it.
A spec initiates discussion. A working feature begs for approval. When I ship something built, the discussion naturally anchors to my solution and subtly converts a group of collaborators into a group of reviewers, much to my chagrin. It happens through politeness because the thing looks done, the Founder made it, and the path of least resistance is a thumbs up with a few suggested tweaks to show you’ve been paying attention. No one has to be dishonest for this to happen, which is what makes it hard to catch.
A finished looking prototype is still only one person’s opinion, and it can be wrong the same way a memo or a spreadsheet with optimistic assumptions in row 14 can be wrong. Ten iterations of my taste is still my taste. The polish is not evidence of anything except the medium.
So two things have to change in this new environment.
The sender needs to ship their thesis with their artifact. The actual why: Here’s the problem I want to solve, here’s what I believe after working on a solution, here are the things I’m unsure about, here’s what I know I don’t have right yet, and here are the parts to ignore completely. A prototype carries the what and strips out the why, so if you don’t surface the reasoning on purpose your team inherits a conclusion with no argument attached and starts guessing at which details were decisions.
Everyone has to be obsessed with solving the core problem instead of taking pride in a particular solution, and that goes double for whoever authored it. The room has the harder job, because it’s psychological. I built the thing, I’m the only one who went on the journey to get there, and now my ego COULD be sitting inside a feature.
The perverse part is that these artifacts are cheaper to throw away than anything we have ever made in the past, and they can feel like the hardest to throw away, because they look expensive. A doc I labored over for a week deserves to be defended more than a feature I generated in an afternoon even though it has rounded corners and it works when you click it.
The author has to be the one who says “this is a draft that happens to run.”
One more objection worth taking seriously. Plenty of people are bad at this, just as plenty of people are bad at writing and bad at spreadsheets. Nobody concluded that Excel was a failed technology because most of the world cannot build an elegant model. We now have a new medium with the same distribution of talent.
Nobody ever lived in the blueprints
The uncomfortable part is that this is not just a Startup problem. It shows up in almost every job where somebody pays a professional for expertise.
Somebody who hires a general contractor wants a house. They do not want schematic drawings, a soil report, load calculations for the HVAC, or a permit set, and yet the industry rule of thumb puts those soft costs at about a quarter of the project budget. One dollar in four buys paper instead of a house.
A company that hires a strategy consultant wants a decision it can defend. What arrives is a deck. Weeks of interviews and benchmarking get compressed into eighty slides that exist so a room full of executives can argue their way to a choice that fits on one page.
Hire a law firm and what you want is the deal closed and the downside covered. What you pay for is a memo, a redline, a diligence report, and a closing binder nobody opens twice.
None of those clients ever wanted the artifact. They tolerated it because there was no other route to the thing they were actually buying, and because the people producing it were the only ones who could.
This is why AI lands as a threat rather than a tool for so many people. It is not coming for the house, the decision, or the deal. It is coming for the paper in the middle, and for millions of skilled professionals, the paper in the middle is the deliverable they’re getting paid for. When the proxy disappears, so does the need for the human expertise (and the billing) attached to it, which is why every industry is going to defend their status quo much harder than they should if they could analyze the benefits without bias.
What this does to the org chart
Almost nothing meaningful gets built by one person, so the tax you pay is the tax of moving an idea out of one head and into several others without losing its shape in transit. Every format we ever invented, from the memo to the Gantt chart, has tried to lower that tax, and none of them have fully fixed the underlying problem, which is that a description of a thing is not the thing.
The gap is finally closing. A meaningful slice of middle coordination exists purely to translate between proxies, and translation stops being valuable when everyone can hand over the real object.
So stop writing the description and build the thing. Use it enough to find out what you actually think and when you’re ready, send something semi-working to your team along with your reasoning.
What none of us knows yet is what happens to the people whose entire job was the description. I’m optimistic, but not because things always work out. I’m optimistic because the people who spent twenty years producing the proxy are the ones who know what good looks like, and knowing what good looks like is the only part of this innovation cycle that AI can’t hand you. The architects who can drive the models are worth more than they were last year, not less. The risk isn’t that the work disappears. It’s that delivering it in this environment requires experience AND a willingness to adapt to technological innovation in order to remain competitive.