Over the past few weeks, we’ve seen the public markets melt down over the fears of AI. Public software companies are now trading at 3x revenues (far from the 10x exit multiple assumption in your financial model!).
Several prominent software names are trading down 50%+ YTD. Figma, a public markets darling post IPO, is now down 75% since their IPO in August 2025.
All of this is generally attributed to AI. A few software companies posted weaker than expected earnings, and suddenly everyone decided that these companies were all going to get killed by AI-native startups.
At this point, every time Anthropic releases a new blog post, billions in stock market value gets wiped out.
But while the public markets noise is deafening, a handful of private companies are quietly thriving. Many founders see this as an opportunity to eat the lunch of many of these flailing public competitors.
Building in 2026 is fundamentally different from the “SaaSacre” of the last decade. If the ZIRP era was about buying market share at any cost, this era is about the Barbell of Success: extreme technical differentiation on one end, and extreme shipping velocity on the other. We see these two archetypes as the two blueprints AI startups are leveraging to get ahead.
Anything in between is “Death Valley.”
Archetype 1: The Luminaries (Technological Differentiation)
These are the companies solving the “really hard” problems that require world-class expertise. Their moat isn’t just a head start; it’s the sheer complexity of their research and engineering, which creates a barrier to entry that capital alone cannot bridge. If you actually own your own cutting edge research and breakthroughs, nobody can vibe code their way into copying you.
Some obvious examples include:
-
Anthropic (Agentic Orchestration): In their 2026 Agentic Coding Trends Report, Anthropic highlights a shift from single agents to coordinated multi-agent teams. They’ve moved beyond simple “coding assistants” to systems that can work for days with minimal human intervention, something that most experts thought was still years away. Anthropic continues to separate themselves from the pack when it comes to building complex, code-heavy agents. By consistently delivering on the bleeding edge, your insights compound and others struggle to match your efficacy. There’s a reason Claude Code is probably the most beloved DevTool in a generation.
-
World Labs (Spatial Intelligence): Led by Fei-Fei Li, World Labs is assembling the greatest computer vision talent on the planet to recreate our world in 3D. While most are stuck in 2D text, they are building the “Marble” platform to generate interactive 3D virtual worlds governed by physics and geometry. This requires a technical bar so high—often involving the world’s leading experts leaving cutting-edge universities—that it creates an immediate, defensible moat. When there’s literally nobody else out there who can do what you do, you don’t have to worry about being undercut.
-
Cohere (The Enterprise Specialist): Cohere has doubled down on being the “cloud-agnostic” choice for highly regulated industries. Their focus on Retrieval-Augmented Generation (RAG) and agentic AI through their “North” platform allows enterprises like Oracle and RBC to deploy models in private or air-gapped environments. Cohere has leaned hard into the “messy” industries that are the most difficult to deploy in: government, telecom, airlines, banking, healthcare, etc. By focusing on customers that everyone else is afraid to serve, you develop unique, proprietary know-how that sets you apart.
The simple answer here is to focus on solving harder problems than anyone else. If you’re willing to tackle the hardest, most complex problems out there, you’re likely going to be alone in the room.
There’s a reason why companies like SpaceX and Palantir have captured massive amounts of value over the last decade (hint: monopolies tend to be lucrative!)
Archetype 2: The Sprinters (Shipping Velocity & Last-Mile Delivery)
On the other end of the barbell are the companies that move faster than the incumbents can even schedule a meeting. Their moat is being closer to the customer problem than anyone else, creating a feedback loop that makes their “last-mile” delivery indispensable.
-
ElevenLabs (Audio): No longer just “the voice cloning company,” ElevenLabs has pivoted aggressively into ElevenAgents, always-on conversational AI designed to replace brittle phone menus and shallow chatbots. This was a bold bet to replace much of their legacy business, even as the legacy business was growing extremely well. While most companies fear the innovators dilemma of replacing their old products, ElevenLabs saw where the puck was going and wasn’t afraid to burn the boats and reboot their core offering on the fly as technology evolved.
-
Harvey (Legal): Harvey is obsessed with the “last mile” for lawyers. For years, many pundits have claimed that Harvey’s product isn’t unique, and that they’d just be killed by OpenAI or Thomson Reuters. In spite of this, Harvey has effectively written the playbook on vertical AI, and they’ve done so with a combination of world class customer support (including on-site training workshops), product sense / attention to detail (“good enough” product feedback has never been good enough), and executing on tight feedback loops (often shipping customer feature requests in the same week they were submitted). What could’ve easily been dismissed as a GPT-wrapper in the early innings has become a formidable business with deeply valuable proprietary data.
-
OffDeal (Investment Banking): OffDeal is the quintessential “full-stack” AI company. They didn’t just sell software to bankers; they became the bank. By using custom agents to automate the tedious analyst work—generating CIMs and financial models in hours instead of weeks—they’ve dramatically improved the customer experience, with a fraction of the headcount. By becoming their own customer (not interpreting pain points second hand, but actually building an offering for their own pain points), they’ve been able to build a custom tech stack in-house that is specifically tailored to their exact workflows and the specific of their industry. There’s no way any generalist offering can ever match those insights.
To be explicit, the common threads here are:
-
Ship weekly (if not daily). All these founders are working 996 (or more)
-
Reassess your product roadmap monthly (even quarterly is too slow)
-
Be more customer obsessed than anyone else (e.g., get constant product feedback, iterate rapidly, trade notes with thought leaders, literally sit down and watch how your customers work, etc.)
-
Don’t be afraid to burn the boats and disrupt yourself if the writing is on the wall (do not let yourself fall victim to your own innovators dilemma!)
Archetype 3: The “Death Valley” (The No-Man’s Land)
This is the most dangerous place to be in 2026. This bucket is filled with companies that aren’t pushing the technical envelope, but also aren’t shipping weekly or talking to their customers every single day.
-
The Lack of Iterative Feedback: If you aren’t shipping at high velocity, you aren’t building the iterative feedback loops required to move from a generic “GPT-wrapper” to a mission-critical tool. Perfect is often the enemy of good: spending too much time tinkering can lead to being left behind in this environment.
-
The Commoditization Trap: Without proprietary research or extreme customer obsession, you are stuck in a race to the bottom. You’re serving up the same results as everyone else because you’re just a layer on a frontier model without the “last-mile” polish.
-
The Valuation Albatross: Many of these tech companies are marked at unrealistic valuations. If you’re not growing 200% YoY (at least), raising another venture round may not be in the cards.
Long story short, we think startups need to pick a lane and hammer on their advantage. If you aren’t the smartest guy in the room in 2026, you’d better make damn sure you’re the fastest.
Conclusion: Pick Your Side
The AI revolution is no longer a theoretical debate about AGI; it is a ground war of execution. The “barbell” is real: either be the one inventing the future of intelligence or the one delivering its value to the customer’s front door faster than anyone else.
Whatever you do, don’t get stuck in the middle. Death Valley is getting crowded, and the water is running out.