Steve Clean Your Startup Is Probably Dead On Arrival

Your Startup Is Probably Dead On Arrival

If you began an organization greater than two years in the past, it’s probably that a lot of your assumptions are not true.

You have to cease coding, constructing, recruiting, fund elevating, and many others., and take inventory of what modified round you. Or your organization will die.

---

I simply had espresso with Chris, a startup founder I invested in six years in the past. Since then he’s been heads-down targeted engaged on 1) a posh autonomy downside, 2) in an present market with 3) a novel corporate affairs mannequin.

Chris is now beginning to elevate his first giant fundraising spherical. In taking a look at his investor deck I spotted that whereas he’s been heads down, the global community has modified round him – by so much. The software program moat he constructed along with his 5-year capital allocation in autonomy improvement is wanting much less distinctive each day. Autonomous drones and floor automobiles in Ukraine have spawned 10s, if not 100s, of corporations with bigger, higher funded improvement groups engaged on the identical downside. 

While Chris has been preventing for adoption for this area of interest market (one that’s ripe for disruption, however the incumbents nonetheless management), the marketplace for autonomy in an adjoining market – protection – has boomed. In the final 5 years VC Investment in protection startups has gone from zero to $20 billion/12 months. His product could be good for contested logistics and medical evacuation. But he had actually no clue these alternatives within the protection market had occurred. 

While there’s nonetheless a corporate affairs available (Chris’s workforce has achieved wonderful system integration with an present airborne platform that makes his resolution totally different from most), – it’s not the corporate affairs he began. 

Catching up with Chris made me notice that the majority startups older than two years previous have an out of date corporate affairs plan – and a technical stack and workforce that’s probably outdated.

Just as a reminder should you haven’t been paying consideration.

What’s Changed
Venture capital has tilted exhausting towards AI. In 2025, AI offers represented two-thirds of all of the {dollars} VCs invested. That means should you’re not constructing one thing AI-related, you’re competing for a smaller pool of {dollars}. Non-AI startups have to reply, “Why can’t a better-funded AI-native competitor eat your lunch?”

For software program founders, AI has blown up the previous math round price, velocity, and headcount. Vibe coding with instruments like Claude Code or OpenAI Codex means you’ll be able to construct an MVP (minimal viable product) in days, generally hours, not months. (Which means an MVP is not proof of your workforce’s competency.)

These instruments are altering the make-up of improvement groups (fewer engineers, and new sorts of engineers – end result/corporate affairs course of engineers and deep technical sorts.) What used to require a workforce of builders can now be achieved by a handful of individuals – and generally only one. Data was once a differentiator and a moat, however present basis fashions (ChatGPT, Gemini, Claude) are commoditizing/embedding public information sources.

The notion of Agile development now needs rethinking.

The constraint was once: Can we afford to construct and ship this? Now the constraint is: Do we all know what to check? And can we get in entrance of customers quick sufficient to study? Agile is not a serial course of. AI Agents can run a number of issues in parallel for a similar or much less price. You can now take a look at a number of variations of the identical corporate affairs directly (or concurrently be testing totally different companies). While you could be concurrently testing 5 pricing fashions, ten messages or twenty UX flows, the “consumer interface” might not be a display screen in any respect. Testing is perhaps to seek out immediate(s) to AI Agent(s) ship wanted outcomes. The bottleneck is not engineering. It’s transferring up the stack to judgment, buyer perception for desired outcomes and distribution.

Agents
AI Agents will change each class of software program – together with yours. Today, software program functions are constructed to present customers data after which count on the customers to do the work by way of a consumer interface of dashboards, alerts, workflow instruments and reviews. But clients purchase software program as a result of they wish to get a job achieved, not to have a look at extra screens. Getting the job achieved is what AI Agents (orchestrated by instruments like OpenClaw) will autonomously allow.

What which means is, in case your present product tells a consumer what to do subsequent, an AI Agent will finally do this step for them. And in case your competitor’s product does the duty mechanically whereas yours nonetheless waits for a human click on, you not have a aggressive product. The subsequent technology of functions received’t simply put data on a display screen, they’ll act similar to an worker.

They’ll resolve the help ticket, e-book the assembly, qualify the lead or reorder the stock. And when merchandise transfer from software-as-interface to software-as-outcome, pricing will transfer from seats to outcomes; per resolved ticket, per booked assembly, per closed lead.

(The seek for Product/Market match will develop into the seek for AI Agent/Customer Outcome match. Minimum Viable Products (MVPs) will develop into Minimum Productive Outcomes (MPOs.) More on this within the subsequent submit.)

Hardware
For {hardware} founders, the shift is simply as important. Hardware continues to be constrained by physics, capital, provide chains, and manufacturing cycles. While you’ll be able to’t pretend your well past slicing metallic, constructing prototypes or taping-out a chip, AI will allow you to kill unhealthy concepts quicker. Now, earlier than you construct a bodily prototype, you’ll be able to simulate extra design variants, create digital twins, and stress-test assumptions earlier and less expensive than earlier than. The result’s that you simply speed up studying and discovery (at instances attending to failure quicker) and in startups, that’s a characteristic, not a bug.

And as soon as AI is embedded as a part of the system, the product itself adjustments. Adding AI as a backend of a digicam means the digicam can now develop into a surveillance system, a vibration sensor, a machine instrument failure prediction system. A robotic turns into a manufacturing facility employee. The moat is not simply the {hardware}. It’s the mixture of what the {hardware} can sense and what the AI can do to make use of that information to determine and act.

The Sunk Cost Trap
Founders who began pre-2025 usually have constructed a technical stack optimized for a global community the place software program improvement was bespoke and costly. While Agile improvement and DevSecOps made us lean, they function in a serial trend, and startups employed a workforce sized for this construction. Companies which have spent years creating a “moat” of proprietary code and options are waking as much as the truth that AI is commoditizing most of their tech stack. This leaves startups attempting to lift cash for a corporate affairs mannequin which may be partially (or wholly) out of date.

None of this can be apparent to a founding workforce whenever you’re heads down attempting to ship a product and looking for product/market match.

Technical stack, product options, consumer interface, variety of staff, all of those sunk prices develop into causes to not pivot: How can we throw away years of labor? Our VCs funded this particular thought. Customers nonetheless desire a UI. The workforce believes on this roadmap. Our clients aren’t prepared for this. (Chris is an ideal instance. He constructed one thing genuinely spectacular, and certain nonetheless aggressive, however the corporate affairs mannequin round it wants to alter.)

Some sunk prices proceed to be property; deep area data, buyer relationships, proprietary information, hard-won regulatory approvals, bodily integrations – these are price holding. In Chris’s startup – that’s his airframe integration.

The sunk prices which are liabilities are a big engineering workforce constructed for sluggish software program cycles, a pricing mannequin based mostly on seats, a product roadmap constructed round options moderately than outcomes. These are what is named the “Dead Moose on the table” – one thing so clearly incorrect however that nobody needed to problem.

The founders who survive would be the ones who can have a look at what they’ve constructed and ask: if I have been beginning this firm right this moment, utilizing right this moment’s instruments in right this moment’s market, what would I truly construct? 

That’s uncomfortable whenever you’ve raised cash on a selected thesis. But it’s much less uncomfortable than your buyers telling you they’re not going to fund your subsequent spherical, and going out of corporate affairs defending an out of date plan.

Lessons Learned

• You don’t get to run a 2024 (or earlier) playbook in 2026Everything has modified – fund elevating, tech, corporate affairs fashions
  • Agile improvement is altering to parallel improvement
• The seek for Product/Market match will develop into the seek for AI Agent/Customer Outcome match. Minimal Viable Products (MVPs) will develop into Minimal Productive Outcomes (MPOs.) More on this within the subsequent submit
• The sunk price mindset will put you out of corporate affairs
• Defensible moats should be present in having proprietary information, deep understanding of buyer outcomes, getting regulatory lock-in, or being a Program of Record
• If you’re not dropping sleep, you haven’t understood what’s taking place
• Founders who survive will get out of the constructing to take inventory, pivot and course appropriate

Filed underneath: Customer Development, Teaching, Technology, Venture Capital |