Father of Silicon Valley entrepreneurship Steve Blank: In the AI era, startups older than two years should consider starting over.

Author: Steve Blank

Translation: Shen Chao TechFlow

Shen Chao Guide: The author of this article, Steve Blank, is very famous in Silicon Valley’s startup scene, known as the “Father of Lean Startup.” He wrote “The Four Steps to the Epiphany” and is the originator of the Customer Development methodology.

Eric Ries’s “The Lean Startup” is developed based on his theories. He has taught entrepreneurship courses at Stanford, Berkeley, and Columbia University, and the NSF’s I-Corps program is also built on his methodology.

Recently, Steve Blank had coffee with a founder he invested in and discovered that the other party had been working diligently for six years without realizing that the outside world had already changed.

This inspired him to write this article, with a very direct core message:

If your company has been established for more than two years, your business plan is probably already outdated. AI is reshaping development speed, team size, pricing models, and barriers to entry. Founders still running on the 2024 script are unlikely to make it to the next funding round.

For readers who are starting a business or are interested in tech and venture capital circles, firsthand observations from across the ocean are worth a read.

Below is the full translation.

If your company has been around for more than two years, many of your original assumptions are likely no longer valid.

You need to pause whatever you’re doing—whether it’s coding, product development, hiring, or fundraising—and first see what’s happening around you. Otherwise, your company will die.

Anxiety Triggered by a Coffee

I just had coffee with Chris. Chris is a founder I invested in six years ago, and since then he has been working tirelessly on:

1. A complex autonomous system problem,

2. In an existing market,

3. Using a unique business model.

Chris is now preparing for his first large-scale funding round. I looked at his investor deck and found one problem: during these years of hard work, the outside world has changed dramatically.

The autonomous system software barrier he spent five years building is becoming less and less unique. Ukrainian autonomous drones and ground vehicles have spawned dozens, even hundreds, of companies with larger teams and more funding doing the same thing.

Chris has been fighting for customer adoption in his niche market (which indeed needs disruption, but the old players still hold the power). Meanwhile, a neighboring market’s autonomous technology demand has exploded—namely, defense.

In the past five years, VC investment in defense startups has skyrocketed from zero to $20 billion annually. His product is perfectly suited for logistics and medical evacuation in contested environments. But he knows nothing about these opportunities in the defense market.

Chris’s team has indeed done impressive system integration (deeply integrated with an existing flight platform, making his solution different from most competitors), and there is business, but it’s no longer the business he initially envisioned.

After talking with Chris, I realized: most startups older than two years have business plans that are already outdated, and their tech stacks and team configurations are probably behind the times.

If you haven’t looked around recently, here’s what you’ve missed.

What Has Changed

VC money is heavily flowing into AI. By 2025, AI projects will account for two-thirds of total VC investments. This means if you’re not working on AI-related stuff, you’re competing for a smaller pool of funding. Non-AI startups must answer a key question: why can’t a better-funded, AI-native competitor directly take your market?

For software founders, AI has completely rewritten the old formulas of cost, speed, and manpower. Using tools like Claude Code or OpenAI Codex for vibe coding, an MVP (Minimum Viable Product) can be built in days or even hours, no longer needing months. This also means MVPs no longer prove your team’s capability.

These tools are changing the makeup of development teams: engineers are fewer, and the types of engineers are evolving, with distinctions like “business process engineers” and “deep tech engineers.”

Tasks that once required a full development team can now be done by a few people, sometimes just one. Data used to be a differentiator and moat, but now foundational models (ChatGPT, Gemini, Claude) are commoditizing open data sources.

Caption: Model T vs Ferrari

The concept of Agile development itself needs rethinking.

The old bottleneck was: can we afford to build and release this product? Now, the bottleneck is: do we know what to test? Can we reach users fast enough to learn? Agile is no longer a serial process. AI agents can run multiple things in parallel at the same or even lower cost. You can test multiple versions of the same business, or different business directions simultaneously. You can run five pricing models, ten marketing messages, twenty UX flows at once. And “user interface” might no longer be a screen; the test goal could be: find the prompt that makes the AI agent deliver the expected result.

Caption: From UI to AI Agent

The bottleneck has shifted from engineering capability to judgment, insight into customer expectations, and distribution ability.

AI Agents Will Rewrite All Software Categories

AI Agents will change every software category—including yours.

Today’s software applications work like this: present information to users, then let users operate via dashboards, alerts, workflow tools, and reports. But customers buy software to get a job done, not to look at more screens. Truly completing the work will be autonomously achieved by AI Agents (orchestration through tools like OpenClaw).

What does this mean?

If your product now tells users “what to do next,” AI Agents will eventually do that step for the user. If a competitor’s product automatically completes the task while yours still waits for a user to click, you will lose your competitiveness.

Next-generation applications won’t just display information on screens; they will act like employees: resolving tickets, booking meetings, screening sales leads, restocking automatically. As products shift from “software as interface” to “software as outcome,” pricing will also shift from seat-based to outcome-based: per ticket resolved, per meeting booked, per lead closed.

(The pursuit of Product/Market Fit will turn into the pursuit of AI Agent/Customer Outcome Fit. MVP will become MPO (Minimum Deliverable Outcome). I will elaborate on this in the next article. ))

Hardware Won’t Escape Either

For hardware founders, the changes are equally dramatic. Hardware still obeys physical laws, capital, supply chains, and manufacturing cycles—you can’t bypass cutting metal, prototyping, or chip fabrication. But AI allows you to eliminate bad ideas faster. Now, you can simulate more design variants before physical prototypes, create digital twins, and test hypotheses earlier and cheaper. The result is accelerated learning and discovery (sometimes faster to failure), and in startups, faster failure is an advantage, not a flaw.

Once AI is embedded as part of the system, the product itself changes. Add AI backend to a camera, and it becomes a surveillance system, vibration sensor, or machine failure predictor. Robots become factory workers. Moats are no longer just hardware itself but what the hardware perceives and what AI can do with that data.

Sunk Cost Trap

Companies founded before 2025 often have tech stacks optimized for an expensive, customized software world. Agile and DevSecOps make us leaner, but they operate serially, and team sizes are structured accordingly. Companies that spent years building “proprietary code and feature moats” are discovering that AI is commoditizing most of their tech stack. This puts startups in a tricky position: their business models may already be partially or fully obsolete.

When you’re heads down building products and seeking Product/Market Fit, these changes may not be visible.

Tech stacks, features, UI, staff size—all these sunk costs can become reasons not to pivot: how can we throw away years of work? Our VCs invested in this direction. Customers still want UI. Teams believe in this roadmap. Our customers aren’t ready yet.

(Chris is a typical example. He built something truly impressive, likely still competitive, but the business model around it needs to change. ))

Some sunk costs are assets: deep domain knowledge, customer relationships, proprietary data, regulatory approvals painstakingly obtained, physical integrations. These are worth keeping. Chris’s flight platform integration falls into this category.

The real liabilities are: large engineering teams built for slow software cycles, seat-based pricing models, product roadmaps built around features rather than outcomes. These are the so-called “dead moose on the table”—obvious problems, but no one wants to call them out.

The founders who survive are those who can look at what they’ve built and ask: if I started today, with today’s tools, in today’s market, what would I do?

When you’re already funded for a specific direction, this question is uncomfortable. But compared to investors saying they won’t fund the next round and you shutting down with an outdated plan, this discomfort is nothing.

Summary

You cannot run the 2024 (or earlier) script on the 2026 track. Funding, technology, and business models have all changed. Agile development is becoming parallel development.

The pursuit of Product/Market Fit will turn into the pursuit of AI Agent/Customer Outcome Fit. MVP will become MPO (Minimum Deliverable Outcome).

The sunk cost mindset can lead to failure.

Defensible moats may still exist in: proprietary data, deep customer understanding, regulatory lock-in, or becoming a formal procurement (Program of Record).

If you can still sleep peacefully, it means you haven’t fully grasped what’s happening.

The founders who survive will step out of their offices, see the situation clearly, pivot, and course-correct.