Focus on the "boring" industries still reliant on phones, fax machines, and outdated workflows—such as law, construction, and elder care—where massive AI transformation opportunities lie.

Guest: Greg Isenberg

Podcast Source: Greg Isenberg

Original Title: 23 AI Trends Keeping Me Up at Night

Release Date: April 2, 2026

Key Takeaways

In this episode, I’ll walk you through my complete list of AI trends and opportunities that keep me awake at night—literally. From the “One-Hour Company Stack” to ambient businesses, vertical AI, agent economies, and the real security threats I see, I’ll share my perspective on what I believe is the most asymmetric window in startup history. I’ll also reveal the framework I use to think about “what to build,” “what to avoid,” and why acting now matters more than waiting for stability.

Highlights Summary

One-Hour Company Stack

Old vs New Startup Timeline

Ambient Business & Autonomous Companies

Agent Economy Timeline

Agent Hires Agent

Vertical Agent Map

Vertical AI vs Vertical SaaS

Opportunities in Vertical Markets

SaaS Pricing Evolution

Outcome-Based Pay

SaaS Graveyard

Scarcity Flip

Premium / Human-Made Products

Experience Economy Explosion

Founder-Agent Fit

Ghost Teams

Micro-Monopolies

Agent Security Threats

Agent Injection vs Phishing

Permission Management

The Window is Closing

Why Asymmetric?

Build in Public

【Trend 1: One-Hour Company Stack】

Greg Isenberg:

Hello everyone! Today I want to talk about the AI ideas that keep me up at night. I’ve compiled a list full of exciting opportunities, concerning challenges, and actionable concepts you can try right away. If you stick with me until the end, you might just start losing sleep over these topics too. Maybe they’ll spark inspiration or deepen your understanding of our current technological and social landscape—and maybe even expose some of the concerns I’m wrestling with.

I want to share the ideas that truly consume my thoughts—the ones that excite me, energize me, and feel incredibly compelling. Perhaps you’ll feel the same way. If you’re listening to this, I suspect you’re someone who thrives on spotting opportunity—one who spends 90% of their time thinking about new possibilities and only 10% of their time afraid of the unknown—but still actively seeking those sparks and insights that drive progress.

First, let’s dive into a concept I’ve been obsessing over: the “One-Hour Company Stack.” Imagine having an idea, quickly writing some code via vibe coding, building a simple landing page, and integrating a payment tool like Stripe—you could attract your first customers almost instantly. The sheer possibility alone is staggering! And better yet, you can directly access platforms like ideabrowser.com to pick from already validated ideas and implement them using your favorite vibe coding tools. This capability is mind-blowing—within a single day, you can launch a new company.

From my perspective, I’m constantly asking how to maximize this power. I don’t want to just launch one company and spend six months validating it. I want to create a culture or mechanism where I can simultaneously launch multiple ventures, testing different ideas—whether targeting the same user group or multiple distinct markets (we’ll get into user groups later). The entire “One-Hour Company Stack” concept keeps pushing me to explore how best to leverage it.

【Trend 2: Old vs New Startup Timeline】

Greg Isenberg:

The second trend keeping me up at night is the contrast between the old and new startup timelines. This is closely tied to the first. In the past, launching a company typically followed this path: You had an idea, hired a few developers (if you could find the right people), spent months building the product, and—if everything went smoothly—released a minimum viable product (MVP) around month three. Then you’d go live, perhaps post on Product Hunt, generate some buzz, and finally see your first revenue around month twelve.

But by 2026, this timeline has been completely disrupted. You might have a new idea at 9:00 AM, pick a validated concept from Idea Browser, start coding with a vibe coding tool at 9:15, finish the product by 9:45, secure your first customer by 10:00, and begin iterating based on feedback before lunch. Some might ask: “How is this possible? Isn’t this just a bunch of immature code written via vibe coding?” But several key factors explain why this is now feasible.

First, you can now use an agent engineering platform—not just a basic vibe coding environment. Tools like Claude Code, or competitors such as Codeex and Google AI Studio, have evolved into extremely powerful systems. These advancements allow us to rapidly build fully functional solutions. Simply relying on these tools enables tasks previously considered impossible—this alone is exhilarating.

Second, you need an email list, an audience, or a group of early adopters to actually attract users. Without that, finding customers becomes incredibly difficult. However, if you're already building distribution channels and have some traction there, you gain a massive advantage. This is one of the main reasons I’ve been losing sleep lately—I’m deeply focused on how to use AI to optimize distribution channel development.

Moreover, I’m reflecting on the contrast between traditional timelines and the new accelerated pace. AI enables us to achieve goals that once required massive resources and time, now at dramatically lower costs and faster speeds. This shift is redefining our understanding of time and efficiency—and opening unprecedented opportunities for founders.

【Trend 3: Ambient Business & Autonomous Companies】

Greg Isenberg:

Another thing keeping me up at night is the concept of “ambient business” or “autonomous companies.” An ambient business is one that requires minimal human intervention—or nearly none at all. These operations are run entirely by AI agents that autonomously handle market monitoring, opportunity discovery, task execution, and customer support. As a founder, you’d only need to check in every few days to monitor performance and progress.

I believe we’re heading toward a future where these ambient or autonomous businesses can achieve annual revenues in the seven- or eight-figure range. The idea is captivating. While we’re still in early stages—many autonomous company software solutions feel rough—I firmly believe this direction is correct. I love calling this trend the “arrow of progress,” which is propelling us toward a world where you don’t need to micromanage every detail because robust governance mechanisms ensure agents act in alignment. This space holds immense commercial potential.

【Trend 4: Agent Economy Timeline】

Greg Isenberg:

The agent economy timeline is another trend that keeps me up at night. From 2009 to 2015, we were in the App Store era—users downloaded apps and manually operated them to complete tasks. From 2015 to 2024, the API economy emerged as developers integrated various APIs to build complex services. I believe from 2025 to 2030, the agent economy will officially arrive. In this era, AI agents will dynamically discover and hire other agents, rendering the fixed team model obsolete.

Within this context, I see a massive startup opportunity: creating a Glassdoor-like platform for AI agents. How do we establish reputation systems for agents? How do we decide which agent to hire? If someone could build a platform akin to Mold Book—which Meta acquired for ~$200 million—focused on AI agent social networks, it would be revolutionary. I know it sounds distant, but I’m convinced it will happen.

【Trend 5: Agent Hires Agent】

Greg Isenberg:

I recently saw a forecast—Gartner, I believe—that by 2030, 20% of business transactions will be agent-to-agent or machine-to-machine. This raises a critical question: How do we build startups that convert existing internet products into agent-native versions? According to projections, this market could reach $52 billion by 2030. Currently, there are over 31,000 agent skills available—but most are low quality. Thus, developing higher-performing, smarter agent skills represents a huge opportunity. I’m genuinely excited by the potential.

Imagine a scenario where agents hire other agents—CEO agents, sales agents, dev agents, marketing agents, etc. Recently, I completed a tutorial using Paperclip, which centers exactly on this concept. Paperclip is an open-source technology whose core idea is transforming traditional organizational structures into serverless functions: agents automatically break down tasks into subtasks and close themselves after completion.

This isn’t just about designing prompts using the “Jobs to Be Done” framework—it’s about hiring agents to manage other agents and execute work, much like hiring employees. This approach is not only innovative but also carries enormous commercial potential.

【Trend 6: Vertical Agent Map】

Greg Isenberg:

According to Y Combinator’s forecast, over 300 unicorn companies will emerge in vertical AI within this decade—proving the massive opportunity in vertical software. Much like Constellation Software, which owns over 500 vertically focused SaaS companies across high-margin processes in education, defense, and more, these seemingly “boring” sectors are actually highly profitable.

Now, similar opportunities are appearing in vertical AI. If you’re listening to this, ask yourself: What’s your unique competitive edge? What vertical domain are you truly skilled in? Those who dig deep into the vertical agent map will have enormous upside. YC typically focuses on major lanes like insurance, real estate, logistics, elder care, legal, healthcare, and sales—but my advice is not to jump straight into these crowded arenas. Instead, pick a specific niche, start small, then scale gradually. These large markets attract massive capital, while smaller niches offer less competition and greater opportunity.

【Trend 7: Vertical AI vs Vertical SaaS】

Greg Isenberg:

I’ve been pondering a key question: What’s the difference between vertical SaaS and vertical AI? Vertical SaaS typically captures only a small portion of enterprise spend. You sell software licenses; the tool must be operated by humans; and typical business scale ranges from $10M to $100M (with exceptions). Vertical AI is fundamentally different—it directly targets enterprise labor costs. You’re building “agent-as-software,” where companies buy your product to replace tasks formerly requiring human hires.

Therefore, the vertical AI market is vastly larger than vertical SaaS. You must think in terms of selling outcomes and results—because agents are doing the actual work. I believe the average commercial value of vertical AI will far exceed that of vertical SaaS. SaaS captures IT budgets; vertical AI replaces labor costs—and labor costs are ten times larger than IT budgets.

【Trend 8: Opportunities in Vertical Markets】

Greg Isenberg:

Which “boring but high-potential” vertical markets deserve attention? Answer: Industries still operating with legacy tools—those relying on phone calls, fax machines, and outdated workflows. Think insurance (still using 30-year-old actuarial tables), law, logistics, elder care, government, accounting, construction. Dig deep into these spaces to uncover hyper-niche submarkets. If I were choosing, I’d avoid highly regulated, high-barrier sectors like direct government sales—too many hurdles. The more mundane the sector, the more niche the market, the greater the potential. It’s a perfect entry point.

【Trend 9: SaaS Pricing Evolution】

Greg Isenberg:

SaaS pricing models are undergoing significant transformation. Historically, SaaS pricing was seat-based—e.g., $50 per user per month—adopted by virtually all major SaaS companies. Yet this is one reason many SaaS firms have seen dramatic stock declines recently: some lost 50–60% of their valuation, dropping from 12x revenue to just 4x. Two key drivers: declining seat demand and investor fear that anyone can now rapidly build alternatives via vibe coding.

Thus, SaaS pricing is evolving through three phases: seat-based → usage-based (“pay for what you consume”) → gradually shifting toward outcome-based pricing (“pay per result delivered”). The core driver is the rise of agents, which actually perform work. Gartner predicts that by 2030, 40% of enterprise SaaS will use outcome-based pricing, while seat-based pricing will drop from 21% today to 15%.

Where’s the opportunity? How can we start building outcome-based business models now? This is a fertile field. Early movers gain first-mover advantage. Whether via cold email outreach, content on social media, or newsletters, people are eager for this innovation—your product could go viral.

【Trend 10: Seat-Based vs Outcome-Based Pay】

Greg Isenberg:

The shift from seat-based billing (e.g., $100/month per seat, regardless of usage) to outcome-based pay is incredibly compelling. Many of us have felt this—no names, but my own company, Late Checkout, pays thousands monthly for certain SaaS tools, yet I often wonder: Are we getting real value?

Now, enterprises can pay based on actual results—e.g., $1.50 per resolved ticket, or only for delivered outcomes. Mature players like Zendesk have already adopted this model, and data shows 83% of AI-native SaaS companies have transitioned to outcome-based pricing. I firmly believe someone could build a $1B company simply by converting traditional SaaS into outcome-based pricing. Helping others make this shift is a massive opportunity—but why help others? You can build your own outcome-driven startup instead.

【Trend 11: SaaS Graveyard】

Greg Isenberg:

I believe a “SaaS Graveyard” is inevitable. Which SaaS companies will be eliminated? I think general-purpose CRM tools will be hit first—excluding giants like Salesforce or HubSpot, which are already pivoting. But if you’re a generic SaaS player not adapting, your survival space will shrink fast—agents can outperform traditional tools.

Additionally, basic analytics dashboards face bleak prospects—AI can generate insight-rich reports on demand. Template markets will become harder to compete in—AI generates highly customized templates instantly. Calendar management tools face disruption—agents natively manage calendars. Basic customer chatbots are being replaced by advanced AI systems, reducing their long-term viability.

【Trend 12: Scarcity Flip】

Greg Isenberg:

What will retain value in the AI era? Answer: tools, infrastructure, and data patterns that successfully transition into AI agent-driven vertical workflows. We’re experiencing a “scarcity flip”: AI rapidly commodifies generic content, foundational design, data entry, and routine analysis, eroding their value. In this context, what becomes scarce and premium?

Myself and many others discussed this on Twitter. The conclusion: value shifts from “execution” to “judgment”—including creative judgment, craftsmanship, and unique physical experiences.

I’m currently incubating projects around this. I believe this is a massive opportunity. By 2026 and beyond, “authentic weird ideas” will be incredibly valuable. Why? Because despite LLMs’ strengths, they struggle with truly “weird” ideas. Everyone has unique life perspectives and experiences—combined with proprietary data, these become the most valuable assets in an AI-driven world.

【Trend 13: Premium Products】

Greg Isenberg:

What defines a “premium” product in the AI era? My answer: content created 100% by humans. You may have heard Porsche’s recent “100% human-made” ad campaign, even launching a “No AI” contest. I believe luxury brands will increasingly embrace “human-made, no AI involvement” as a core ethos—akin to organic certification in food. “No AI” could become a new quality signal. This warrants deep thought—and may unlock similar opportunities elsewhere.

【Trend 14: Experience Economy Boom】

Greg Isenberg:

Another important layer within premium products is the “AI-augmented, human-led” model. In this paradigm, human involvement becomes the hallmark of premium quality in the AI era—combining human creativity and taste with AI’s efficiency. Fully AI-generated services risk becoming commoditized and trapped in price wars.

That’s why I’m especially interested in incubating IRL (In Real Life) projects. As the digital world becomes infinite and AI-generated content floods in, scarcity naturally shifts toward physical presence and shared human experiences. Thus, venues like karaoke bars, escape rooms, immersive theater, co-working spaces, and live concerts are central to the experience economy. It’s booming—full of excitement—and one of the reasons I can’t sleep.

【Trend 15: Founder-Agent Fit】

Greg Isenberg:

Another fascinating emerging concept I call “Founder-Agent Fit.” Reflecting on my past startups—especially after moving to Silicon Valley—everyone talked about “Founder-Market Fit.” The core question: Do you deeply understand your customers and market? As a founder, do you possess unique insights into the space? For example, if you’re building a social network for college students, are you still a student?

Now, we’re entering an era of “Founder-Agent Fit.” As a founder, you must be able to orchestrate and command an entire AI agent team to achieve your goals. This is analogous to a film director: the director doesn’t operate cameras, act, or compose music—but extracts the best performance from actors and crew. In the future, those “actors” will be AI agents. Thus, “Founder-Agent Fit” will become a core skill for the next generation of founders. I find this shift fascinating and highly promising.

If you can design and manage AI agents within a niche market, unlocking their full potential, you’ll have a massive competitive advantage—directly related to earlier discussions about Paperclip and zero-human companies.

【Trend 16: Ghost Team Org Structure】

Greg Isenberg:

In the future, company “Team” pages may become “Ghost Team” pages—showing only a handful of real employees, with the rest filled by AI agents: Sales Agent, Content Agent, Support Agent, etc. You can even name them, give them personalities, generate avatars, and simulate video calls or voice messages—creating near-indistinguishable human-like collaboration experiences.

As a founder running a holding company and incubating new ventures, I believe more holding companies will emerge. Because AI-native agent businesses will dominate, and companies can efficiently operate multiple ventures in similar or identical niches using ghost teams.

【Trend 17: Micro-Monopoly Business Logic】

Greg Isenberg:

Kevin Kelly proposed the “1,000 True Fans” theory. But in the AI era, I believe 100 true fans are enough. AI agents drastically reduce operational costs, so just 100 paying customers can sustain a profitable business. Since agents efficiently replace human labor, you can deliver high-value services—say, $1,000 or $500 per month per client. With just 100 customers, you can build a highly profitable enterprise. Even lower fees work, since your operational cost approaches zero—possibly just one person.

This low-cost, high-efficiency model will spawn countless micro-monopolies. For example, if you have 5,000 highly active niche followers, you can build a custom app in 48 hours. Via email lists or newsletters, you might easily find 100 customers paying $50/month. Using agents to run the business, you could earn $60,000/year profit—already substantial. And you can replicate this model to incubate more such ventures.

Of course, acquiring the first 100 customers is key. Thus, building an efficient content production and distribution system is essential. Even without an existing audience, you can buy traffic—though it reduces margins—this remains a viable strategy.

【Trend 18: Agent Security Risks】

Greg Isenberg:

While I’m optimistic about AI’s future, one issue keeps me up at night: the attack surface of AI agents. You’ve likely heard of threats like prompt injection, context window poisoning, malicious MCP services, agent manipulation, privilege escalation, and corrupted training data. By granting agents broad access, we’re opening doors to potential vulnerabilities. If I said I wasn’t concerned, I’d be lying. I believe malicious incidents are inevitable—and current cybersecurity tech lags far behind agent evolution. This risk deeply unsettles me.

Palo Alto Networks recently documented real-world agent injection attacks. If even top-tier security firms like Palo Alto warn of widespread risks, I trust their assessment completely.

【Trend 19: Agent Injection vs Phishing】

Greg Isenberg:

How should we view the relationship between agent injection and traditional phishing? Around 2010, phishing targeted humans clicking malicious links—defense relied on human judgment. Despite this, global losses still reached billions annually. Today’s agent injection is far more complex: hiding instructions to deceive AI agents, primarily targeting context windows and webpage content. Because agents are highly autonomous, this becomes a critical vulnerability.

I believe agent injection will cause far more damage than traditional phishing. When agents have system access and autonomy, poisoning their context window becomes a new, dangerous attack vector. I’m certain we’ll see many such malicious events. But this also creates a massive opportunity: building specialized cybersecurity software for agents. Startups focused on agent security will be a vital frontier.

【Trend 20: Agent Permission Management】

Greg Isenberg:

When using AI agents, we must seriously consider their permissions. Specifically: What resources can they access? Can they access your files, emails, calendar, or even bank accounts? Already, users have granted agents direct access to bank accounts—e.g., “There’s $5,000 here—help me transact.” What can agents remember? Can they store conversation logs, personal data, or business information? What actions can they perform? Can they send emails, shop online, modify code, or delete data? And crucially: Who can they share information with? Can they exchange data with other agents or third parties?

Given this, we must emphasize the concept of “digital hygiene.” Just as we periodically audit app or website permissions, we should regularly review agent permissions—ideally quarterly. For instance, I sometimes realize certain SaaS tools requested unnecessary access, so I disable them. I believe we’ll eventually apply similar practices to agents to ensure digital safety.

【Trend 21: The Golden Window for AI Entrepreneurship is Closing】

Greg Isenberg:

We’re currently in an era where construction costs are nearly zero. AI agents can handle most work; many niches remain untapped; and user acquisition costs are relatively low. But I don’t believe this window will last forever. That’s why I feel urgent and driven. I estimate this golden period will last about 12 more months. Competition will grow, prime niches will be claimed, and some tools will become overcrowded. Within the next 24 months, the window will narrow significantly. Founders who act now will build moats through data accumulation, network effects, brand equity, and trust.

Many wait for the market to “stabilize,” but it never truly does. Rapid change is the norm. In this age of infinite opportunity, every day matters.

【Trend 22: Why the Opportunity is So Asymmetric】

Greg Isenberg:

Today’s opportunity window is profoundly asymmetric. All you need is an API key, a few well-crafted prompts, a tweet, and a niche audience of 100 to 5,000 people to build a 24/7 operation with 95% gross margins—especially for agent-centric businesses. Even if margins drop to 70%, 80%, or 60% over time, these remain exceptional models. With compounding distribution mechanics, these companies can run efficiently with almost no staff—or minimal staff.

【Trend 23: Build in Public】

Greg Isenberg:

I believe we’re living in the most asymmetric era for starting a company. While some argue against “building in public,” I still believe the benefits far outweigh the downsides—especially when your audience is also your potential customer base. By openly sharing your product development, the community can participate in decisions, helping shape direction. The most exciting part of the AI era is launching feature updates in just one to five days. This rapid iteration turns users into co-builders, dramatically increasing trust and distribution efficiency—a powerful growth flywheel.

Moreover, I believe “fork businesses”—borrowing, adapting, optimizing, and innovating existing models—will become common. Like forking a GitHub repo, in a world where copying is easy, attracting community involvement and making them feel part of the creation process will become a critical moat.

In short, this is an exhilarating era of building, though rapid changes can feel overwhelming. But as long as you take that first step, make small daily progress, and accept that you can’t master every AI tool, you can keep moving forward in this era of immense opportunity. It’s an incredible time! Let’s push forward together. See you next time—thanks for listening!