one builder, no agency

For a scoped software project, think a customer portal, an integration between two systems or an internal dashboard, one experienced builder with AI is often faster than an agency in 2026, at a fraction of the cost. The work that five years ago took a project lead, a frontend developer, a backend developer and a tester is now done by one person, with AI doing the execution and the human as final editor. Whether that is possible has been answered by now. The question that remains is about continuity: what happens if that one builder drops out, and you cover that risk with a handful of concrete agreements I spell out further down.

One thing up front. I write this piece wearing two hats: I am such a solo builder myself. So read it as the story of someone with a stake in the outcome, and check me against the sources. That is exactly why I spend half of this essay on the case against.

what shows up in the numbers

The shift toward building smaller is measurable. The share of American startups with a single founder rose from 23.7 percent in 2019 to 36.3 percent in the first half of 2025 (Carta, 2025). Investors remain wary meanwhile: solo companies made up 30 percent of the 2024 cohort but raised only 14.7 percent of the capital in priced funding rounds. So the market believes in the model, with reservations. Those reservations are about precisely the subject of this essay.

At the same time, AI-native software companies are showing how much work can come out of one person. Research by investor Redpoint Ventures, cited by Forbes, puts revenue per employee at AI-native startups at 2 to 4 million dollars, against roughly three hundred thousand at the average listed SaaS company (Forbes, 2026). Coding platform Lovable hit 400 million dollars in annual revenue in early 2026 with 146 people. And the story of Base44, built by one founder in four months and sold for 80 million dollars within half a year, I already told in the new builders.

The best-known story is more extreme, and it comes with a warning. Telehealth company Medvi is said to have done some 401 million dollars in revenue in 2025 with two people on payroll, the founder and his brother, and claims to be on track for 1.8 billion in 2026 (New York Times, 2026). That last figure is the founder’s own projection, not an audited annual number. Analyst firm Forrester promptly put caveats next to it: two people on payroll rarely means two people doing all the work; behind stories like these usually sits invisible labor from freelancers and vendors (Forrester, 2026). The same Forrester analysts report that when companies announce layoffs “because of AI”, nine times out of ten there is no mature, tested AI system ready to take over that work (HR Executive on Forrester, 2026).

The sober reading: the solo unicorn is an exception that attracts media attention. The underlying shift is real. Small teams, and increasingly individuals, deliver work that until recently took a department.

what can a solo developer with ai do that used to take a team?

One builder with AI now covers the whole trajectory: design, frontend, backend, integrations, tests, documentation and delivery. In my own work that looks like this: I put multiple AI agents to work in parallel on different parts, and my day consists largely of reviewing, steering and deciding. The typing is delegated. The thinking and the responsibility emphatically are not.

The research on this is more precise than the hype, and it pays to dig into it for a moment. Faros AI analyzed telemetry from over ten thousand developers in 1,255 teams: teams with high AI adoption merge nearly twice as many pull requests, but review time rose by more than 90 percent alongside it and ate up most of the gain (Faros AI, 2025). In a team, the time AI saves relocates to coordination: more work means more handovers, more review, more alignment. That is where the structural advantage of working alone sits. For me, those coordination costs move into the review, and that review lives in one head. No handovers, no queues.

The honest story also includes the setback. In a controlled experiment by research institute METR, sixteen experienced open-source developers using AI tools worked 19 percent slower on 246 real tasks, while beforehand they expected to be 24 percent faster (METR, 2025). That study covered experienced developers in large, existing codebases, so you cannot paste the outcome onto every project. The lesson stands: the feeling of speed is unreliable, and whoever works alone has no colleague to puncture that feeling. I guard against it by planning and billing on delivered steps, and by tracking per job what it actually cost. If an AI route turns out slower than building it myself, I want to see that in my own numbers, and sometimes I do.

The biggest pitfall sits one layer deeper still: the demo is 20 percent of the work, and why so many AI projects die between pilot and production is something I worked out separately in the production gap.

what is the risk of one builder without a team?

The biggest risk of a solo builder is continuity: if that one person drops out, your project or system stands still. That goes for me too. Software teams have called this the bus factor for decades: how many people have to be hit by the proverbial bus before nobody understands the system anymore. For a solo builder, that answer is one by definition.

AI deepens this risk rather than shrinking it. The critical knowledge now also sits in the way someone directs, corrects and distrusts an AI system, and that kind of skill is harder to document than code. A practitioner analysis by MindCTO sums it up sharply: you cannot write a work instruction for intuition (MindCTO, 2026). That is a blog, not academic research. It does match what I see myself: someone else can take over my code; transferring my way of working demonstrably takes more effort.

Two sober Dutch data points come on top. About 71 percent of independent contractors (zzp’ers) carry no disability insurance (ZZP Pulse, 2026), so solo entrepreneurs already run an elevated personal continuity risk as it is. And according to research by the bank ABN AMRO, 46 percent of Dutch SMBs experience partial to very high dependence on American tech vendors (ABN AMRO, 2026). That second risk I share completely: my way of working leans on AI models from a handful of American providers. Whoever hires me hires that dependency too.

If a solo builder waves these risks away, that in itself is a reason not to work with him. The right response is to name them and cover them contractually.

when is an agency or a permanent team the better choice?

Choose an agency or a team of your own as soon as the system becomes business-critical and needs continuous maintenance, hard guarantees or a lot of coordination. Concretely, I see four situations in which I refer a small-business owner to an agency. One: you need 24/7 availability with SLAs and penalty clauses; one person cannot keep that promise. Two: it is multi-year development with multiple departments and vendors plugging in; then coordination is the product, and agencies are built for that. Three: you operate in a heavily regulated environment, healthcare or finance for instance, where audits and segregation of duties are required. Four: you want your own people to learn to build and maintain; then you are buying guidance, not delivery.

In those cases an agency sells exactly what a solo builder by definition lacks: replaceability of people. That is not an admission of weakness from the solo builder and not an extravagance from the agency. They are different products for different situations.

What you do not need to do in any case: buy a months-long agency program, or put someone on payroll, for a first scoped project of a few weeks. Start small, look at what is actually standing after the first delivery, and let the form of the collaboration grow with the importance of the system.

the six agreements that cover the solo risk

If you hire one builder, set these requirements up front. A good solo builder proposes them unprompted.

1. Ownership from day one. Code, prompts, configuration and documentation live in your environment: your repository, your accounts, your domain. You can revoke the builder’s access, never the other way around.
2. Documentation as a delivery requirement. Every delivery includes a decision log (why it was built this way) and a short video walkthrough. No documentation means no acceptance and no payment.
3. A tested handover. Agree that any other developer can get the system running within one working day using only the documentation. Have that genuinely tested at least once, by a friendly IT contact for example.
4. Standard building blocks. Require mainstream technology that thousands of developers know. An exotic stack only the builder understands turns every handover into a rebuild.
5. Maintenance and exit terms on paper. Who handles updates and outages, what that costs per month, and what happens if the builder is out for a longer period, including a replacement or handover party named by name.
6. Pay per delivered, working step. Then your maximum damage on dropout is one step, and you force the builder to test his own feeling of speed against what is actually standing.

the dutch context: solo contracting is shrinking, and not because of ai

Anyone who thinks AI is causing a wave of solo builders in the Netherlands is looking at the wrong chart. The number of independent contractors fell by 62,000 in 2025 to around 1.2 million, and Statistics Netherlands (CBS) attributes that decline to the tightened enforcement of the rules on false self-employment since January 1, 2025 (CBS, 2026). The contraction is sector-specific too: strong in home care and childcare, while the number of independent contractors in hospitals actually grew. For you as a client, the relevant question is therefore a different one: can the builders who remain structurally handle more than they could three years ago? That is what the numbers in this piece do point to, and it changes the trade-off between agency and solo builder for good.

My position is this. For scoped software and AI projects in small business, one builder with AI, provided the six agreements above are on paper, is the better buy than an agency: shorter lines, lower cost and one person who oversees the whole and personally signs for it. As soon as your system becomes business-critical and guarantees outweigh speed, a team belongs around it, and I say that to my own clients too. That is how I work myself, on fikst among other things.