23 AI Trends keeping me up at night
TL;DR
The “one-hour company stack” compresses startup building from months to hours — Greg says a 2026 flow could be idea at 9:00 a.m., product by 9:45, first customer by 10:00, using tools like Claude Code, Codex, Google AI Studio, Stripe, and a distribution channel like an email list.
Ambient businesses are the real prize: agent-run companies with almost no daily human input — He’s fixated on businesses where agents monitor markets, handle customer service, execute tasks, and eventually drive seven- or eight-figure outcomes while the founder mostly checks in.
Vertical AI is bigger than vertical SaaS because it sells labor replacement, not software seats — Greg contrasts old per-seat SaaS with outcome-based AI, arguing vertical AI taps labor P&L directly, with YC predicting 300+ unicorns in vertical AI this decade across niches like insurance, logistics, elder care, legal, and construction.
AI is creating a scarcity flip where execution gets commoditized and judgment becomes premium — Generic content, design, dashboards, and templates get cheaper, while human-made craft, original weirdness, proprietary data, and physical experiences like karaoke bars, escape rooms, and immersive theater gain value.
Small audiences can now support real businesses because agents crush operating costs — He updates Kevin Kelly’s “1,000 true fans” to “100 true true fans,” giving the example of a 5,000-person niche audience converting 100 customers at $50/month into roughly $60,000 profit with an agent-run micro-business.
The scariest trend is the agent attack surface — Greg worries prompt injection, poisoned context windows, malicious MCP services, permission escalation, and agent-to-agent manipulation will outgrow classic phishing, citing Palo Alto Networks and warning that cybersecurity hasn’t caught up.
The Breakdown
From months-long startups to the one-hour company
Greg opens with the thing he can’t stop thinking about: the “one hour company stack.” His point is blunt — you can grab an idea from ideabrowser.com, vibe-code it, slap on a landing page and Stripe, and get something real into the market almost instantly. He contrasts the old startup timeline — months to hire devs, build an MVP, and maybe get revenue by month 12 — with a new one where you can build by 9:45 a.m. and iterate by lunch.
Distribution becomes the real bottleneck
He’s clear that speed alone isn’t enough: if you don’t have an audience, first customers are still hard. That’s why he keeps coming back to distribution as the hidden edge — email lists, social reach, or AI-assisted audience building. The subtext is that coding is getting cheap fast, but access to buyers is still precious.
Ambient businesses and the rise of the agent economy
Next he gets excited about “ambient businesses,” meaning companies that run with near-zero daily human input. He imagines agents monitoring markets, finding opportunities, handling support, and only needing occasional founder oversight — with seven- and eight-figure businesses eventually running this way. He ties that to a broader timeline: app store era from 2009–2015, API economy from 2015–2024, and now the 2025–2030 “agent economy,” where agents hire other agents and someone should build the “Glassdoor for AI agents.”
Vertical AI: boring industries, giant outcomes
Greg then zooms into vertical AI, where YC predicts 300-plus unicorns this decade. His advice isn’t to chase giant categories head-on, but to wedge into sub-niches inside insurance, real estate, logistics, elder care, legal, healthcare, or construction — especially boring sectors still powered by phone calls and faxes. The core argument is that vertical SaaS captures IT budget, while vertical AI replaces labor, making the market much larger.
The pricing shift from seats to outcomes
A big section is about how AI breaks traditional SaaS pricing. Greg notes major SaaS stocks are down 50–60%, partly because investors expect fewer seats and worry buyers can just vibe-code alternatives. He sees the world moving from per-seat pricing to usage-based and now to outcome-based pricing — like paying $1.50 per resolved support ticket, with companies like Zendesk already moving this way and Gartner projecting 40% of enterprise SaaS to shift by 2030.
What dies, what gets premium, and why weird wins
He sketches a mini “SaaS graveyard”: generic CRM, basic dashboards, template marketplaces, scheduling tools, and simple support products all look vulnerable if they don’t evolve. In their place, the premium layer shifts toward human judgment — original taste, weird thinking, physical experiences, and human-made craft. He uses Porsche’s 100% human-made campaign as a signal that “no AI involved” could become a luxury certification, while AI-assisted but human-led becomes premium and fully AI-made drifts toward commodity.
Founder-agent fit, ghost teams, and the 100 true true fans model
One of his more memorable reframes is moving from founder-market fit to “founder-agent fit.” He says founders will look more like film directors, orchestrating fleets of agents rather than doing the work directly, and he imagines future team pages filled with a couple humans plus named AI employees. That feeds into his economic thesis: in the AI age, it may only take 100 true true fans — not 1,000 — because agents slash costs enough for small, niche, high-margin businesses to work.
The thing that actually freaks him out: agent security and the short window to build
After all the optimism, Greg says the biggest fear is the expanding agent attack surface: prompt injection, poisoned context windows, malicious MCP services, permission escalation, and compromised training data. He compares it to phishing, but worse, because agents can have system access and autonomy, and he argues founders will need “quarterly agent cleanses” to review permissions the way they review app access today. He closes with urgency: build costs are near zero, audiences are underpriced, and there may be a 12- to 24-month window before niches crowd up and the best moats — data, brand, network, trust — get claimed.