Product Field Work All writing
WritingFounders

Starting a company, then and now

In 2000, starting a software company meant raising millions for servers and licenses before a single customer saw anything. Today a founder can put a working product in front of customers within days. The hard part hasn’t disappeared. It has moved, to getting noticed, earning trust, and keeping customers who can switch in an afternoon.

The problem

Every playbook founders use was written for a world where building was the expensive, slow part. The lean startup existed to stop teams from spending a year building something nobody wanted. Venture capital existed partly to pay for engineers. Moats were often just the head start of having built something hard.

When a competent founder with coding agents can build most of that in weeks, those assumptions break. Not all at once, and not equally everywhere, but enough that a founder starting today needs a different map.

$20Mwhat Marc Andreessen said it took to start an internet company around 2004, for routers, servers, Oracle licenses and paid distribution
36%of new startups on Carta were solo-founded in 2025, up from about 24% in 2019. On Stripe Atlas it was 63% in Q2 2026
14.7%of priced-round cash went to solo-led startups founded in 2024, even though they were 30% of new companies. Teams still get funded
How I compare the eras
  1. Capital needed before a customer can try the product
  2. Time from idea to a real customer test, the first turn of the lean loop
  3. Team size at launch
  4. What the hardest problem actually was
Why I wrote it

I start things. I’ve felt the difference between building a product over quarters and building one over a weekend, and I’ve also felt how little the weekend version matters if nobody knows it exists or trusts it with their data. This is the founder’s side of the same shift the rest of this site describes.

The ideaBuilding is no longer the moat. Learning from customers faster than anyone, reaching them before anyone else, and earning a kind of trust a weekend clone can’t fake: that’s where a founder’s time now goes.

1Five eras of starting up

What it took to get a first product in front of customers

Pick an era. Dollar figures for 2000–2010 are from investors at the time; the rest are my estimates, and they vary a lot by product.

2The lean loop, re-timed

Steve Blank and Eric Ries gave founders one loop: build something small, measure how customers respond, learn, repeat. For twenty years “build” was the long leg. Now it’s the short one. Blank himself put it plainly this year: the 2024 playbook doesn’t work in 2026, and development is becoming parallel rather than sequential.

How long each leg of the loop takes

Arc length is time. Build shrinks dramatically. Measuring and learning from real customers shrinks much less, because customers don’t speed up.

BuildMeasureLearn

The loop is no longer limited by how fast you can build. It’s limited by how fast you can get honest answers from customers.

That’s why the conversations still matter most. In Blank’s own Stanford class this spring, eight teams spoke with 978 potential customers, beneficiaries and regulators. Agents can research a market in an afternoon; they can’t tell you whether a buyer will actually change how they work.

3The copycat clock

The flip side of building fast is that everyone else can too. A feature that once gave you a year of lead now gives you weeks, sometimes days, before someone ships the same thing with two more features and half the price.

After you launch a new feature, who else has it?

Illustrative. Drag the slider to move forward in time and watch lookalikes appear around you in each era.

2016 0 lookalikes

2026 0 lookalikes

4One person, and then who?

The one-person company is real. Solo founding has roughly doubled over a decade, and a solo-built product, Base44, sold for $80 million. Anthropic’s CEO has put high odds on a one-person billion-dollar company arriving soon.

But the data also says something quieter: building the first version got dramatically cheaper; scaling a company didn’t. Recruiting, selling to enterprises, covering many domains under growth pressure, and earning trust are still people problems. The question for a solo founder isn’t whether AI is enough to start. It’s who to add first.

Your first five hires

Switch between a typical early SaaS team and the one I’d build now. The engineering work doesn’t vanish; it moves into the founder’s workflow and the agents.

5Attention is the new scarcity

Posting on social media is free and instant for everyone. That’s exactly why it’s hard to be heard. Software is heading the same way: when anyone can ship a polished product, the product alone stops getting attention. The founders who win already have an audience, a community, a partner channel, or a place inside the tools customers use every day.

This shifts where a founder spends the first year. Less on building, which agents now carry; much more on distribution, from day one. Building an audience in public, owning one channel deeply, getting found in AI search, and partnering with whoever already has the customer’s attention are no longer “later” work.

6Why should anyone trust you?

A buyer looking at ten products that do the same thing has a new question: which of these was built last weekend, and which will still be here, secure and supported, in three years? Trust used to come mostly from being a known vendor. Now it has to be built on purpose, in layers.

The trust stack

From the foundation up. Tap a layer for what it means in practice.

7Keeping customers who can leave in an afternoon

When a rival can copy your product, add personalization and charge half, customers become fickle. The old defenses weaken: features, code, shallow integrations. Others get stronger. Here’s how I’d rate them, then and now.

What still defends a software business

My judgment, informed by the founders and investors cited below. The hollow marker is 2016; the solid one is today.

WeakNeutralStrong

The common thread: the strongest defenses are the ones that get better with time and can’t be generated. A product that learns each customer’s patterns, becomes where their work actually lives, and is backed by people they trust gets harder to leave every month. A feature list gets easier to copy every month.

8The founder’s playbook now

First two weeksLearn faster than anyone
  • Have agents map the market, competitors and complaints in a day.
  • Then talk to 30 real buyers yourself. That part doesn’t compress.
  • Build three rough working versions and put them in front of the best ten.
First two monthsEarn the right to charge
  • Ask for money early; a pre-sale is the only validation that counts.
  • Pick one narrow customer and one channel, and own both.
  • Start the trust basics now: security, support, a public changelog.
First yearBuild what can’t be copied
  • Become where the customer’s work lives, not a tool beside it.
  • Let the product learn from each customer, with their permission.
  • Raise, if you raise, for distribution and trust, not engineering.

Choices

Distribution from day oneThe first hire I’d make isn’t an engineer. It’s someone who owns how customers find you.
Trust as a featureSecurity, uptime, support and transparency are part of the product, and they’re the part a clone can’t fake.
Depth over breadthOne narrow customer served completely beats ten served with features anyone can copy.
Where I’d push back on the extreme version

A one-person unicorn is possible; it will also be rare. Teams still attract most of the funding. Code that was fast to write can be slow to secure and maintain, which YC’s own partners warned about as AI-written codebases took over their batches. Blank rethought his class so students would build businesses “rather than AI slop.” And in regulated markets, the time it takes to earn a buyer’s trust hasn’t shrunk at all. Speed of building is a real advantage. It just isn’t the one that decides who wins.

Sources

  1. Marc Andreessen, 2014, quoted in the Bangor Daily News: about $20 million to start an internet company a decade earlier.
  2. Duncan Davidson, quoted by RingCentral: launch costs from $5 million (2000) to $500,000 (2005) to $50,000.
  3. Carta, Solo Founders Report 2025: 23.7% to 36.3% solo-founded; 14.7% of priced-round cash.
  4. Solo Founders, citing Stripe, June 2026: 63% of Stripe Atlas C corps in Q2 2026 were solo-founded.
  5. Forbes, August 2026: Dario Amodei on the one-person billion-dollar company; Base44’s $80 million acquisition.
  6. TechRound, May 2026: the barrier to scaling hasn’t fallen like the barrier to building.
  7. Steve Blank, Customer Development: the 2024 playbook in 2026, parallel development, and 978 customer conversations in one spring class.
  8. Harvard Business Review, September 2026: AI is changing the rules of entrepreneurship.
  9. TechCrunch, March 2025: a quarter of YC’s W25 batch had codebases 95% AI-generated, and the partners’ caution about reading and debugging it.
  10. Joe Reis, April 2026, Designli’s survey of 100 SaaS founders, July 2026, and Levera Partners, May 2026: feature velocity isn’t a moat; data, workflow depth, distribution and trust are.