Thoughts on the vanity bottlneck
The primary friction preventing widespread enterprise adoption of LLMs is vanity.
I find it ironic that the profession closest to the machine is the one most comfortable with this truth, AI researchers. They see their own work get automated. Literature reviews, hyperparameter tuning, even parts of architecture search. They watch it happen to their colleagues and themselves. No one pretends these tasks required irreplaceable human intuition. They just built tools to do it faster.
This is not a trend across industries. Yet.
Man is the vainest of all creatures that have their being upon earth. - Homer
The pattern I keep seeing. Customer success teams manually handling account escalations. Implementation consultants building the same integrations for the hundredth time. Product managers conducting research calls that could be synthesized from existing data. Sales engineers giving demos that differ only in the company name on the slide deck.
These are companies built on the promise that they know better, and its easier to use them than do it yourself. The recurring revenue model funds an army of knowledge workers doing repeatable, automatable work.
Uncomfortable Truth
I talked with an individual in a leadership position at a Series C company recently. Strong quarter. ARR up, churn down, board happy. When I asked about their AI roadmap, he got honest in a way executives rarely do.
"We've got some experiments running, sure. But honestly my team's performance looks great. Why would I volunteer to prove half of them unnecessary?"
This is where the bottleneck lives. Not in technical capability. Not data quality. Not even the "our use case is different" excuse.
It's vanity.
These companies spent years hiring strategically. Building teams. Developing specializations. They told themselves and their boards that this headcount was their moat. The accumulated knowledge in their customer success org. The intuition of their implementation team. The relationships their account executives built.
And they were right. For a while.
The Relationship Delusion
The most common defense I hear is about relationships. "Our customers buy from us because of the relationship." "We're a consultative sale." "Our AEs are trusted advisors." This is where SaaS companies drift into consulting territory without admitting it.
If your business model depends primarily on personal relationships rather than product value, you're not really a software company. You're a services firm that happens to have some code, if any at all (you'd be surprised how many companies are just an amalgamation of other service companies). People, processes, and technology translates to we own nothing.
Relationships feel defensible. They're warm, human, irreplaceable. But you don't own them. They walk out the door when people leave. They reset when territories get reassigned. They scale linearly at best. Real product value scales differently. It compounds. It doesn't need to be rebuilt with every new customer, every new point of contact, every organizational change.
When I see SaaS companies leaning heavily on relationship selling, I see firms that haven't built enough product value to stand on its own. Now, with AI capable of handling increasingly sophisticated interactions, that gap is showing.
Why Earnings Keep Everyone Honest
When the competitor automates and you don't, margins tell the story the most people won't.
If your competitor runs customer success with 30% of your headcount and maintains the same NPS, your board will ask questions. If they implement new customers in days instead of weeks, your sales cycle suffers. If their product team ships at 2x velocity because they're not doing manual research, you fall behind.
The market will force the decision that leadership can't make internally.
This is where things are headed. Not doom and gloom, but a reckoning. The companies that have been honest about what's automatable will have a head start. The ones clinging to "our knowledge workers are special" will have a rougher transition.
What Actually Matters
If relationships aren't the moat and knowledge work is automatable, what's left?
Taste. Judgment on the edge cases. Knowing which problems are worth solving. Understanding when to override the AI's recommendation because context matters.
These are human skills. But they're different from what most knowledge workers were hired to do.
The companies that figure out how to redeploy their best people toward these higher-order problems will do fine. The ones that keep everyone in their current roles to preserve the org chart will get disrupted by their own customers.
The tough question is coming. Why fund all this overhead when companies that simply wrap LLMs are faster, better, and dramatically cheaper?
Tldr;
This isn't about technology replacing humans. It's about companies being honest with themselves about what their humans are actually doing, and whether it's defensible.
Most knowledge work in SaaS companies isn't. The vanity is pretending otherwise.