Vertical AI Maximalism
Or: why the wedge is dead
The Duomo di Milano. Maximalism was the original maxxing.
Starting a Vertical AI company today with a wedge product is lame. This strategy made sense when software was hard to build, incumbents were friendly, and integrations were reliable. In the last few months, none of those hold true anymore.
But I have been mystified by the stubborn persistence of this short-sighted “wedge thinking.” This is not armchair criticism. I built a wedge company and believed in a version of the strategy for many years.
So let me explain why the wedge is dead, and what should take its place: Vertical AI Maximalism.1
A Brief History of Wedges
To understand the current opportunity in Vertical AI, let’s revisit the period from roughly 2012 to late 2025. That era had three defining traits.
First, SaaS exploded. As industries moved from on-prem to cloud and from paper to software, a generation of Vertical SaaS platforms emerged. Think public companies like Veeva in life sciences, Procore in construction, AppFolio in property management, and many others.
Those platforms went on to create meaningful ecosystems. The core product couldn’t do everything, after all. For many years, the dominant posture of incumbents was to allow API integrations with startups. The model looked a lot like Salesforce’s original AppExchange: if you were a startup, there was a reasonably clear path to plug into the system of record, win some partnership support, and build from there.
Second, building software was hard, expensive, and slow. Startup engineering teams were measured in dozens of people. Teams built roadmaps in quarters. And $10s of millions in venture went to headcount to ship industry tailored, UI heavy CRUD apps.
Land (SF office space), labor, and capital were real constraints. These limited inputs produced “wedge thinking.” If you wanted to compete with an incumbent, you were told to start narrow, earn a foothold, and expand from there. Eventually, you could work your way to the platform. At the time, that logic made sense, and founders built many large companies following this approach, even a few platforms.
Third, and most surprising to me, wedge thinking persisted well into the post-GPT era of 2023-2025. As recently as late last year, many vertical AI startups were still narrow by design, often focused on a single medium such as voice. The growth was impressive across industries like healthcare, logistics, and services. The AI use cases were real and novel. But the product strategy? Still unmistakably wedge-shaped. The best founders in these categories have already started to build full platforms, but must still support their wedge product customers.
What’s Changed
The obvious answer is coding. Since January 2026, Claude and Codex have dramatically compressed the time required to build software. Products that once took years to build can now ship in weeks. Agents also work in production now. We are no longer limited to voice-only AI applications or lightweight “copilots” assisting human workflows.
That matters because workflow software itself is changing. Historically, Vertical SaaS depended on users moving through screens, filling fields, and pushing records through point-click interfaces. In the agent world, the system can execute most (though not all) of that work. The shape of the product changes when the agent, not the user, becomes the primary operator.
What has been less evident than the technology shift is the new game theory in Vertical AI.
Until 2025, many Vertical SaaS incumbents were still willing to run the old partnership playbook: open (enough) APIs, plenty of integrations, and encouragement for ecosystem startups. It was a Vertical SaaS NAFTA. The system worked, albeit imperfectly.
But as Vertical SaaS multiples halved earlier this year, the tone has changed. A number of incumbents have become much more willing to engage in defensive, even anticompetitive, behavior: reducing API access, shortening partnership windows, raising integration fees, and generally making life harder for startups that depend on them.
In the long run, customers will gravitate toward whatever products help them most. Competition is good, and many incumbents will lose ground to emerging startups.2 Let the best products win. But in the short run, this new dynamic changes startup product strategy. Wedges a priori require integrations. Wedges also assume the incumbent’s cooperation, or at least its benevolence. That era is over.
The Comparative Opportunity in AI Services
Another market is emerging at the same time: AI Services. The relevant comparison for vertical AI is no longer just Vertical SaaS, but services companies rethought with AI.
There has been a lot of slop written on this topic lately, and I will save a fuller treatment for another post. But the core point is valid: AI makes possible a new class of services businesses with larger TAMs, better margins, and more defensibility than their non tech enabled predecessors. Some of these companies, if executed well, will generate substantial free cash flow. Some may end up larger in market cap than the Vertical SaaS winners of the last generation.
The challenge for Vertical AI founders is narrative risk. Compared to services, VCs could typecast Vertical AI companies as “too niche.” In some verticals, that advice is directionally right.3 In others, it completely misses the point.
For better or worse, startups are a comparison business. VCs often judge founders against the available alternatives at the moment, with a not-so-secret bias toward where to deploy the greatest amount of capital. Let’s not forget the pace and scale of marketplace VC investments during the 2010s, for instance.
Vertical AI Maximalism
Vertical AI Maximalism is the view that founders should build end-to-end, agent-native platforms that directly replace core industry platforms rather than entering through narrow, integration-dependent wedges. The opportunity is orders of magnitude larger than the previous generation of Vertical SaaS.
But this new reality also requires a new level of founder and product ambition.
The founders who succeed here will have two attributes. First, they must have a combination of: technical ability to push Claude/Codex to their limits and real market insight. Founders need to master the nuances of running ~15 agents in parallel and produce high quality, performant products for critical workflows like accounting, HR, and payments. Second, the most recent Vertical AI founders seemed to be light on competitive drive. But I would wager that this generation of founders will need the relentless intensity of the 2010s marketplace founders. Fortune favors the brave.
The products themselves should be new types of platforms, not AI replicas of incumbent software. The point is to rethink what should be done by humans, what should be done by agents, and what outcome the customer is buying. The best companies will also capture proprietary data that compounds over time.
The scope of Vertical AI Maximalism does not mean founders should build in a cave before sharing their creation with customers. If anything, it demands deeper intimacy with the customers. Founders have to understand the existing system well enough to decompose it, and rebuild it around a different balance between people and agent labor. This approach is not easy, but the best products never are.
The upside is the chance to build Vertical AI companies far larger than the largest Vertical SaaS businesses of the last cycle, and rivaling the AI Services businesses of the current era. Not just better software companies, but massive cash flow businesses that would have been difficult to imagine even a decade ago. Founders should build toward that future, not toward AI papier-mâché replicas of last decade’s platforms.
Wedges are dead. Long live maximalism.
***
Notes
1. By “Maximalism” I mean product abundance and ambition vs. the inherent restraint of wedges. Some art historian and architecture geek turned PM will probably roast me for this, but you get the drift.
2. Some incumbents will also succeed, even thrive. I’ll leave the prognosticating to public market hedge funds.
3. For very small markets, Claude & Codex might mean the ability for companies to build their own tooling for the first time. Results will vary, but many products will be better than the horrendous incumbent offerings on the market today. Constellation Software, take note.