The Shape-Shifter Economy

The Quiet Defection

The “everyone will be unemployed by AI” headline and the “AI changes nothing, just use it as a tool” headline are both wrong in the same way. They treat work as a binary: you have a job, or you don’t. You write code, or a model does.

The actual shift underway is quieter and stranger. A growing number of capable people are not quitting their jobs. They are staying — because health insurance in the United States is effectively bundled with W-2 employment, and no rational person with a family walks away from that in 2026. But inside those jobs, the nature of the work is changing. And outside those jobs, after hours, on weekends, in the margins, something new is getting built.

The individual developer is becoming a shape-shifter. By day: a reviewer, a shaper, a decider, a user acceptance tester. By night: the founder, architect, and sole engineer of a product that would have required a team of five a decade ago.

Why the Day Job Stays

Let’s be honest about the thing nobody wants to say out loud. In the United States, employer-sponsored health insurance is the single most powerful retention mechanism ever invented. It’s not loyalty. It’s not culture. It’s not stock options. It’s the fact that leaving a job means navigating the individual market, COBRA, or a spouse’s plan — each of which is worse, more expensive, or more precarious than what the employer provides.

This has always warped the labor market. It will warp it harder in the AI era. Because now the economic case for quitting has collapsed in both directions. AI makes solo building radically more feasible, which increases the pull toward independence. But AI also compresses the amount of traditional programming labor employers need, which decreases the individual’s bargaining power and makes the safety of a W-2 feel more precious, not less.

The rational response isn’t to quit. It’s to stay, but let the shape of the job change underneath you.

What the Day Job Becomes

The programming itself is the part AI does well. Writing functions from specifications, translating designs into components, wiring up APIs, generating tests, refactoring known patterns — all of this is increasingly the model’s job. The human’s job is everything the model still can’t do reliably: deciding what to build, deciding whether what was built is actually what was wanted, and catching the subtle, domain-specific ways that correct-looking code is wrong.

Four activities dominate the new day job:

• Shaping. Defining the problem with enough clarity that an AI agent can work against it. This is Ryan Singer’s shaped pitch by a different name. It’s the skill Ethan Mollick keeps pointing at: management, not technical expertise.
• Reviewing. Reading generated code critically, understanding it well enough to approve or reject it, and noticing when a test passes for the wrong reason. Review is becoming the primary act of engineering, not the secondary one.
• Building. Not writing every line. Assembling, orchestrating, and wiring the generated pieces together. Making the judgment calls about architecture, privacy, and performance that a model cannot make with context it doesn’t have.
• User acceptance testing. Actually using the thing. Finding the gap between “the code is correct” and “the product is good.” This has always been undervalued labor. It is about to be the most valuable labor a senior engineer does.

None of this is unemployment. All of it is a significant redefinition of what a senior engineer does for a living. The title stays. The paycheck stays. The insurance stays. The activity changes.

The Side Door

Here is where it gets interesting. The same AI capabilities that are compressing traditional programming at work are making it absurdly cheap to ship a real product alone. Not a side-project toy. A real, deployable, revenue-capable product.

A single person with taste, judgment, and a clear appetite can now:

• Shape a product idea in an afternoon
• Ship a working prototype in a weekend
• Run the whole thing on a $20/month hosting bill
• Handle support, marketing, and billing with a handful of integrated tools
• Iterate faster than a traditional ten-person team, because there are no standups, no roadmap committees, and no coordination overhead

The limiting factor is no longer engineering hours. It’s taste, judgment, and focus. Which are exactly the things a senior engineer with a day job has been building for a decade.

So the pattern emerges: keep the W-2 for the insurance and the stable income. Build the side product with the hours AI just gave you back. Let the two reinforce each other. The day job sharpens your judgment on real systems at real scale. The side product gives you full-stack ownership and a direct line to end users. Neither is a distraction from the other. They’re the same skill, exercised on two different surfaces.

Why Small Teams, Not Solo, Will Win the Next Decade

Pure solo is powerful but lonely, and taste benefits from a second set of eyes. The real unit of the next decade isn’t the lone wolf — it’s the tiny team. Two to five people. Complementary skills. No middle management, because there’s nobody to manage in the middle. No project manager, because the tools do the coordination. No traditional handoffs, because the team is small enough that everyone touches everything.

Small teams compound in ways big teams can’t:

• Decisions happen in conversation, not in meetings. A two-person team makes a hundred decisions a day with zero overhead.
• Everyone shapes, everyone reviews, everyone tests. The roles blur because they have to.
• AI is the third teammate. It’s the one that writes the first draft of everything — code, copy, tests, migrations — so the humans can focus on judgment.
• The product stays coherent. Small teams don’t build Frankenstein features because they share a single mental model and update it live.

This is the world my Merrily / Syncopate bet was designed for. Shape Up was always a small-team methodology. Basecamp built it for themselves. It scales up awkwardly and scales down beautifully. And the AI layer — the thing that does the status tracking, progress reporting, and coordination work in the background — is exactly what makes a two-person team feel like a five-person team without the overhead tax.

What I’m Actually Building For

Every product in the MCG portfolio is, in its own way, built for the shape-shifter and the tiny team. Not enterprise. Not consumer at scale. The independent builder and the two-to-five person crew that is about to become the dominant economic unit of software and creative work.

• Merrily — Shape Up for the team of one to five. Betting tables, cycles, hill charts, financials. Everything a tiny team needs to run like a business without hiring operations staff.
• Mulholland — Film production tools for the writer-director-producer who is doing all three jobs. The same AI-shaping-and-review pattern, applied to screenplays, shot lists, and visual development, instead of code.
• Jotto — Desktop daily notes and task management for the individual builder whose moment-to-moment focus is the scarce resource.
• Eddy — Feedback collection and AI classification, so a tiny team can hear its users without hiring a customer success department.
• Rumpus — Events as a solo or duo operation, not as a conference production company.
• Tempo — Privacy-first calendar infrastructure, because the tiny team controls its own data or it doesn’t.

The unifying thesis isn’t “AI replaces people.” It’s AI reshapes who can plausibly run a business alone or with a partner. And the tooling that wins this decade is the tooling that treats that tiny operator as the first-class user — not as a degraded version of an enterprise customer.

The Part Nobody Is Pricing In

There’s a second-order effect that deserves more attention. When a senior engineer keeps the W-2 for insurance, does the day job with half the traditional keystrokes, and runs a side product in the remaining focus, the side product has something almost no VC-backed startup has: patient capital. The runway isn’t measured in months. It’s measured in “how long until I don’t need the day job anymore,” and the answer can be years without burning a single investor dollar.

Products born this way can take their time. They can optimize for sustainability instead of hypergrowth. They can refuse ad-based business models. They can charge fair prices and serve small audiences well. They can practice the privacy-first architecture that investor pressure usually forbids, because there’s no investor to pressure them.

This is a different kind of software economy than the one venture capital built. It’s slower, smaller, more personal, and in aggregate, probably much larger than the VC portion. The shape-shifters in the aggregate will ship more software than the funded startups. They will own more customer relationships. They will collect less data and charge more honest prices for it.

The Punchline

The defining worker of the 2020s software economy isn’t the unemployed engineer replaced by AI, and it isn’t the ten-engineer startup team either. It’s the senior builder who keeps the day job for the insurance, lets the shape of that job drift toward shaping-reviewing-deciding-testing, and uses the focus AI just gave back to build something of their own on the side.

That person needs tools. Not enterprise tools retrofitted with AI chat. Not consumer tools optimized for engagement. Tools built from the ground up for the individual and the tiny team who are doing the work of a company of ten and want to keep doing it that way.

That’s who I’m building for. That’s the whole bet.