Fable and the vertical AI moment

The full impact of the Fable drama is yet to be fully understood. But if nothing else, it will hasten the push towards companies eliminating their dependence on any single frontier model. Having your intelligence effectively shut off by a US government directive is an untenable risk.

As a result, enterprises will be on the search for orchestration layers, advanced routing, eval systems, open source models, and small language models (SLMs). The reality is the movement against “tokenmaxxing” was already happening. We’ve invested behind this thesis a handful of times already and are excited to meet any companies forming part of the “rebel alliance,” as Michael Mignano called it yesterday. But Fable is just going to accelerate this movement.

A second order effect may be the rise of the application layer. Given Fable’s extreme capabilities, this isn’t obvious. But there are a handful of trends converging that make me think there’s a chance the best of the vertical AI companies (and probably some new entrants as well) are about to go on absolutely historic runs.

Here’s what I’m seeing:

First, we’ve already got Fable making CIOs everywhere jittery. They want rock solid, model agnostic infrastructure they can trust. 

Second, Fable made abundantly clear is that building software just really isn’t that hard anymore. It’s all about distribution and deployment. The teams capable of doing this better than others will hold a massive competitive advantage.

Of course, if you’d been paying attention, you would have realized this already. The best vertical AI companies are not those with the best products. They’re the ones who have cracked distribution (a.k.a. the right vertical paired with the right wedge) and deployment. If you dig in with any of these vertical AI winners, you’ll see that superior deployment leverages FDEs not just to implement the product, but to build out the context capture and routing and learning loops and everything else that makes the product actually work.

Third, all this infrastructure is not just valuable because it makes the product perform better, but because it is, by its very nature, cross-functional. That matters more than you might think.

At Airtable, we always talked about wanting to land in cross-functional, "connector" use cases. We loved content calendars, for example, because the editor worked with the product marketing team on launches, the design team on headers, and the marketing team on promotion. It was a great place to land because you built a system everyone used, which made it both stickier, and easier to expand horizontally.

You get that same horizontality for free with AI. If you are a vertical AI company selling customer support tools, providing a superior experience requires massive amounts of data from other teams. You need product roadmaps, help center documentation, bug reports, and sales context to know if the user is currently on an enterprise trial. In other words, to solve even a vertical problem effectively, you must travel horizontally across cross-functional paths to gather new use cases and context.

Fourth, there are huge pools of investor capital that are desperate to invest in AI, and those investors want the biggest story possible. A lot of this capital is chasing generic enterprise-specific versions of this deployment story. (See Poetic, which just announced a chunky round last week, for the latest). But the problem with these is distribution. How do you land consistently if your solution is so generic? If you believe the biggest bottleneck is distribution, and that this sticky horizontal infrastructure layer gets built by FDEs…then you probably see where I'm going.

The vertical AI companies aren’t going to stay vertical for very long.

They’re going to raise gobs of money selling a horizontal story. They’ve already done the hard part. They’re capturing loads of context, they have the trust of their customers and they have FDEs on site. Now they just need more products to sell. To fill in the gaps, they’ll acquire other vertical AI products (acquiring companies requires lots of capital, which in this case, is a good thing!). Or just build it with the latest model in a few weeks. Maybe they’ll even vertically integrate. Why not? Their own routing layer. Their own model strategy. Their own open-source stack. Maybe even their own neocloud.

I’m being a bit tongue in cheek here, but not really. A lot of investors will tell you that they are done with the application layer. Anthropic and OpenAI are going to eat it all, they say. But Fable has shifted the ground beneath us once again, and I think you’d be crazy not to update your assumptions. With that in mind, I’m itching to make some more vertical bets, so if you’ve got something big brewing, let’s chat.

David Peterson

Could the future of software be fluid
How do we get the best of AI without losing the soul of software?

The future belongs to young missionary teams
Why it makes more sense betting on youth in the current moment

The AI-native enterprise playbook
Ten real-time observations on a rapidly evolving playing field

No more painting by numbers
It’s the end of the “SaaS playbook.

ENTERPRISE/TECH NEWS

ELON - Joe Nocera writes about ‘changing the ticker from SPCX to ELON in The Free Press. SpaceX launched their IPO to fanfare and bafflement. It trades at 95x trailing sales, lost $5 billion in 2025, and claims a $28.5 trillion total addressable market that requires a "lunar economy" to justify. Nocera's read: none of that matters, because with Musk, stock value bears almost no relationship to fundamentals. Bloomberg's Matt Levine called it in 2021, things are valuable not based on cash flows but on proximity to Elon Musk. The float is tiny, index funds are forced to hold it from day one, and Musk controls 85% of votes with no independent directors and no meaningful shareholder protections. A fascinating and slightly alarming anatomy of how the Elon premium actually works.

A MODERN FABLE - The US government's order to suspend foreign access to Fable 5 and Mythos 5 followed a viral jailbreak published on X, in which a prolific hacker claimed to have bypassed the model's safety guardrails to extract instructions for cyber exploits, explosives, and chemical synthesis. VentureBeat's practical read on what this means for enterprise AI strategy and why the Defense Secretary labelling Anthropic a "supply chain risk" back in March was an early warning that single-provider dependency is an organisational liability, not just a procurement preference.

HOW TO STARTUP

COMEBACK KIDS - Intercom, now Fin.ai, announced Salesforce’s intention to acquire the business for $3.6bn. Intercom was founded in 2011, having spent years building the customer messaging category they pivoted hard into AI agents under the Fin rebrand. Eoghan McCabe and Des Traynor will stay on, and Fin will fold its multi-channel AI agent technology into Agentforce.

HOW TO VENTURE

THE UNTRAINABLE - Sarah Guo of Conviction shared on LinkedIn / Twitter why she hasn’t fallen into ennui about the current state of Venture. The AI despair thesis runs: every company built on a model is a thin wrapper waiting to be absorbed, and the only surviving value is compute and frontier weights. Guo thinks this is half right. The wrapper layer is being absorbed, but what remains is more durable than the despair suggests. Value slides toward work whose correctness exists only in private data, inside systems that require security reviews, contracts, and earned trust to enter. A smarter model doesn't make private ground truth public, can't hold the licence, and can't be the party that gets sued. The companies that survive are the ones doing the unglamorous work of translating a firm's private reality into something a model can act on.

A LONG, STRANGE TRIP - Ed Sim, founder of boldstart ventures, has been listening to the Grateful Dead and pondering the ‘long, strange trip’ the tech world is on. His blog this weekend, reflected on both the largest IPO in history with SpaceX now worth $2 trillion and the US government's suspension of foreign access to Anthropic's newest models. The latter makes Sim's long-held thesis feel urgent: the opportunity is in applications that work across multiple models, route workloads intelligently, protect proprietary data, manage costs, and keep running when a provider or policy changes underneath them. Government intervention will only accelerate the need for enterprises and founders to own their models and data.

Fixefy was featured in Supply Chain Digital. CRO Yaniv Butel breaks down why procurement teams keep losing visibility over logistics spend and what agentic AI actually does to fix it.

Groundcover
Account Executive (San Francisco)

FalkorDB
Core Developer (Tel Aviv, Hybrid)

Reco
AI Software Engineer (Tel Aviv)