Five Futures for SaaS (and Which One Is Actually Happening) — Q1 2026

This is the first edition of a quarterly series where we track five competing visions for the future of SaaS and assess which one is winning. We’ll update the evidence and revisit our position every quarter.

As of February 2026, the software industry is in the middle of something big.

Salesforce’s stock is down 20% from its peak. ServiceNow, Workday, HubSpot: the whole sector is in a re-rating that shows no signs of reversing.

Goldman Sachs published a report arguing that AI agents will capture more than 60% of software economics by 2030. Chamath Palihapitiya is calling out a $5 trillion “Software Industrial Complex” that he thinks is about to collapse. David Sacks launched Glue as a bet that traditional collaboration software is already obsolete.

And Ben Thompson, who coined “SaaSmageddon,” wrote this month that the industry-wide re-rating is “completely justified.”

Everyone has a theory about what comes next. The problem is that the theories wildly contradict each other.

Some people think SaaS is dead. Others think it just adds an AI button and keeps going. Others think one mega-platform eats everything.

I spend my days helping companies figure out what AI actually changes for their business and what it doesn’t. And what I’ve found is that people get stuck because they’re working from one mental model when the reality is a collision of several.

So let me walk you through the five most common visions for the future of SaaS, explain why each one is partially right and mostly wrong, and then tell you what I actually think is happening. I’ll revisit this every quarter as the evidence shifts.

Future 1: “SaaS Is Dead. Agents Build Everything.”

This is the most dramatic version of the story, and it’s increasingly popular.

The argument goes like this: if an AI agent can build a custom tool for your exact workflow in 20 minutes, why would you pay $50/seat/month for a generic product that does 40% of what you need and 60% of what you don’t?

There’s real evidence behind this. I recently watched a GTM engineer build five custom internal tools in a single afternoon using Claude Code: a bulk ad generator, a LinkedIn scraper pipeline, a landing page generator, a custom analytics dashboard, and an email campaign system.

Each one took about 20 minutes. Each one was customized to his exact workflow. None of them involved opening a SaaS product.

Chamath Palihapitiya has been making a version of this argument. His 8090 incubator is built on the thesis that AI enables solo founders to deliver what used to require a team of 100.

If a two-person startup can build enterprise-grade software, why does the enterprise need to buy off-the-shelf SaaS at all?

It’s a compelling vision. And it gets one thing exactly right: a huge chunk of the SaaS market exists not because the software is hard to build, but because building software used to require a team.

When that constraint dissolves, a lot of products lose their reason to exist.

Here’s why it’s mostly wrong.

Software isn’t just code. It’s also collaboration, shared state, institutional knowledge, regulatory compliance, and reliability at scale.

An agent can build me a personal CRM in 20 minutes. What it can’t build is the network effect that makes Salesforce useful across a 200-person sales team, the compliance infrastructure that makes Epic work in 500 hospitals, or the multiplayer editing experience that makes Figma worth using with a team.

The “SaaS is dead” vision works beautifully for individual power users and small teams doing straightforward workflows. It falls apart the moment you need shared context, regulatory audit trails, or real-time collaboration at scale.

I call these “Mode 2” tools: bespoke software built by agents from APIs. They’re real, they’re growing, and they will kill a specific category of SaaS. But they won’t kill SaaS as a category.

Future 2: “SaaS Just Adds AI Features and Keeps Going.”

This is the incumbent’s favorite story.

The pitch: every SaaS product adds a copilot, a smart search feature, maybe an AI-powered assistant. The product gets better. The per-seat model persists.

Customers pay a bit more for the AI features. Nothing fundamentally changes.

You hear this from SaaS CEOs on earnings calls. “We’re embedding AI across the platform.” “Our customers love the new AI features.” “We’re seeing strong adoption of our copilot.”

And they’re not lying. AI features genuinely improve existing products. Auto-categorization in a support tool saves time. Smart search in a knowledge base is legitimately useful.

But this vision has a fundamental problem.

It treats AI as a feature, not a force. It assumes the industry structure stays the same and AI just makes every product incrementally better.

Ben Thompson’s analysis is devastating on this point. Yes, AI makes it cheaper and faster to build software. But it does that for everyone. Every SaaS company gets more productive, and so does every competitor, and so does every new entrant who might build a replacement from scratch.

The advantage cancels out.

Meanwhile, enterprises are actively cutting SaaS spend to fund AI token budgets. The pie isn’t growing. It’s getting redistributed.

The “just add AI” vision is what you tell investors when you don’t want them to panic. It’s not a strategy. It’s a prayer that nothing structurally changes.

Future 3: “One Agent Platform Eats Everything.”

This is Thompson’s aggregation theory applied to agents.

The logic: in the original version of aggregation theory, zero distribution costs gave power to demand aggregators. Google aggregated content discovery. Facebook aggregated social. Amazon aggregated commerce.

Now imagine the same thing happening to software. If your AI agent becomes your primary interface to all your business tools, whoever controls the agent controls demand. SaaS companies become suppliers, competing on API quality and price, not user experience.

Satya Nadella has been making exactly this play. Microsoft’s strategy is to absorb adjacent business functions into Copilot, making the agent more capable, charging more for it, and compelling other SaaS companies to build connectors that interface with their system.

If Microsoft succeeds, SaaS companies don’t die. They become the backend pipes that Microsoft’s agent calls.

It’s an elegant theory. And it has historical precedent. What Google did to publishers, an agent platform could do to SaaS.

But there are two reasons this probably doesn’t play out in its extreme form.

First, SaaS companies have real moats that content companies didn’t. Switching costs, data gravity, compliance requirements, integration complexity. These don’t dissolve overnight.

When Google aggregated web content, switching from one article to another was free. Switching your CRM is a six-month project. Aggregation will be slower and messier than it was for content.

Second, and more importantly: we don’t know if one platform will dominate. Thompson’s aggregation theory requires a single demand aggregator. If the agent platform layer ends up being competitive, with Claude, ChatGPT, Gemini, and others each holding 15-30% share, no single platform has the leverage to commoditize suppliers.

SaaS companies can play platforms against each other. The open protocol layer (MCP and equivalents) matters enormously here: if agents can call any tool through an open standard, no single agent platform becomes the gatekeeper.

Thompson himself makes this distinction. He predicts “open pipes, closed products”: MCP becomes the highway system, agents are the cars. Nobody owns the highway, but the car companies still capture margin. That’s a competitive platform world, not a winner-take-all aggregation story.

Future 4: “Every SaaS Builds Its Own Agent.”

This one sounds reasonable on the surface. Every SaaS company builds an AI agent that deeply understands its own product. Salesforce builds an agent that’s great at CRM. Notion builds one for knowledge management. Linear builds one for project management.

Everyone gets an AI upgrade. The ecosystem stays distributed.

There’s a problem, though. If every SaaS company builds its own agent, you get the SaaS silo problem back, but worse.

Think about it. Today, when you use five different SaaS tools, at least you, the human, maintain cross-tool context in your head. You know the meeting in Google Calendar relates to the deal in Salesforce relates to the project in Linear relates to the doc in Notion.

Your brain is the integration layer.

Now replace yourself with five separate agents, each one deeply understanding its own product and blind to the others. You’ve recreated silos, except now you’re also context-switching between agents instead of just tabs.

We’ve seen this movie before. In 2016-2018, every company built a chatbot. Users hated talking to 50 different bots. They gravitated back to single interfaces. The same dynamic will play out with agents.

The interoperability problem is the thing that makes this future unstable. Each SaaS company wants their agent to be the primary interface. But users want one agent that works across everything.

That tension pushes the market toward general-purpose orchestrators, which pushes us back toward Future 3: platform aggregation.

The exception is Microsoft, which owns enough of the stack (Outlook, Teams, SharePoint, Dynamics, Azure) that their Copilot genuinely orchestrates across a full work surface without interoperability problems. They don’t need open standards because they own both sides.

Future 5: “The Messy Middle.”

This is the one I actually believe.

None of the first four futures is entirely right, but each captures something real. The actual future is a collision of all of them, playing out at different speeds in different market segments.

Here’s my best understanding of what’s actually happening.

SaaS bifurcates. The industry splits into two distinct shapes. On one side: infrastructure APIs and collaboration platforms that survive because they provide something agents can’t replicate (network effects, scale infrastructure, regulatory compliance, multiplayer coordination). On the other: mid-tier workflow SaaS that gets squeezed from above by platform absorption and from below by custom agent-built tools.

The timeline is different for different segments. For individual power users and small teams, the disruption is already underway. Mode 2 (agents building custom tools from APIs) is happening right now for people who know how to use it.

For enterprise, the timeline is 5-8 years. Data gravity, compliance requirements, and organizational inertia create real friction. The EU AI Act’s human-in-the-loop requirements alone slow the agent-replaces-everything timeline for any company operating in European markets.

Platform competition prevents full aggregation. Three to five agent platforms compete. None dominates. MCP or equivalent becomes an open standard.

This is more like the cloud provider market (AWS, Azure, GCP all thriving) than the search market (Google monopoly). SaaS companies maintain leverage by being cross-platform.

The revenue model stays unresolved longer than anyone expects. Thompson identified what I think is the most underappreciated problem in this entire transition: the agent revenue problem.

When an agent queries TripAdvisor for hotel data, nobody sees ads. The entire advertising and engagement revenue chain gets bypassed. If service providers can’t monetize agent interactions, they stop providing data, and agents become useless.

The proposed solution is micropayments via stablecoins: sub-cent transactions (0.003 cents for a hotel query, 0.001 cents for weather data) that credit cards can’t process because the transaction fee exceeds the transaction itself.

Agents don’t have the psychological friction humans have with microtransactions. They just optimize.

This is the one place where crypto has a clear functional advantage over traditional finance. Not Bitcoin (volatile, slow, expensive), but stablecoins: boring, stable, programmatic.

But the infrastructure doesn’t exist yet. And until it does, there’s a revenue vacuum that slows the entire transition.

Model commoditization shifts value. Open-source models reaching “good enough” (the DeepSeek effect) gradually compress margins at the model layer. This is good for SaaS companies (cheaper to embed AI) but bad for the “arms dealers” thesis that says model makers win regardless.

Value shifts toward whoever controls the customer relationship and the data. Meritech Capital’s analysis argues that this could actually make the SaaS market bigger, not smaller, as AI expands addressable use cases even while individual products face disruption.

What Actually Survives

If the messy middle is the right framing, then the question becomes: which SaaS products end up on which side of the bifurcation?

I’ve stress-tested this across dozens of categories, and the pattern that holds best is what I call the intrinsic vs. instrumental test.

Products survive when the experience of using them is intrinsically valuable. Figma survives because designing in Figma IS the creative process. An agent generating a design file isn’t a substitute for the collaborative canvas.

Products die when the experience is purely instrumental. You don’t want to use a landing page builder. You want a landing page. If the agent produces it directly, the builder’s UX is irrelevant. Same for workflow automation tools, template-based design, basic CRM, simple dashboards.

Here’s the rough hierarchy of what survives, ranked by durability:

Strongest. Deep domain knowledge with regulatory requirements. Epic in healthcare, Veeva in life sciences, financial compliance platforms.

The moat isn’t the UI. It’s the 10,000 edge cases from decades of deployment, the certifications that take a year to earn, and the fact that regulators require provable compliance through auditable processes. LLM judgment doesn’t satisfy auditors.

Strong. Real-time collaboration on complex artifacts. Figma with dozens of concurrent editors on files with hundreds of components.

Collaboration libraries (Yjs, Automerge) commoditize simple co-editing. But the performance engineering at Figma’s scale and complexity is years of work that agents can’t shortcut.

Moderate. Infrastructure APIs. Stripe, Twilio, Cloudflare. These were always “tools agents call.” They survive because they’re already positioned correctly. But “good enough” infrastructure is increasingly achievable with standard cloud primitives, so only the extreme-scale players are truly safe.

Weakest. Network effects (narrower than most people think). Slack’s network effect is real for synchronous human coordination.

But it weakens for async communication. If agents monitor Slack, Teams, email, and Discord simultaneously and present a unified view, the “everyone is already on Slack” advantage erodes.

The Transition Nobody’s Planning For

Here’s the part that bothers me most.

Thompson draws a useful analogy to agriculture. In 1810, 81% of the American workforce was in farming. Today it’s about 1%. Machines replaced human labor, but entirely new kinds of work were created.

He argues the same will happen with AI and software.

He’s probably right about the long-term outcome. He’s dangerously quiet about the middle.

The agricultural transition took 200 years and was accompanied by massive human suffering. The institutions that made it survivable (land-grant universities, labor unions, public education, the GI Bill) took decades to build.

Thompson simultaneously predicts near-term SaaS layoffs (specific, imminent) and long-term job creation (general, philosophical). But he doesn’t spend much time on the bridge between the two.

Who retrains the mid-market SaaS sales team whose product just got absorbed into an agent platform? What does the five-year transition look like for the integration engineer whose job was building connectors that agents now build on the fly?

Every major platform shift has created enormous value for the intermediaries who helped enterprises navigate the transition. Shopify and Stripe for the internet-era commerce shift. Consulting firms and systems integrators for cloud migration. The AI transition equivalent is wide open.

This is the gap we spend most of our time thinking about at Seeko. Not which SaaS products survive and which don’t. That’s an interesting intellectual exercise, but it’s not where the work is.

The work is helping companies figure out which future they’re building for, and then actually building the architecture, automations, and team practices to get there.

It’s not about picking the right prediction. It’s about building systems that adapt regardless of which future arrives.

What This Means If You’re Running a Software Company

I’ll leave you with a few things I’d actually do if I were reading this today, whether you’re a CTO, a founder, or an ops leader trying to figure out what AI changes for your business.

Map your product against the intrinsic vs. instrumental test. Be honest about which parts of your product exist because the UX is genuinely valuable and which exist because building software used to be hard. The first category is defensible. The second is on a clock.

Make your product agent-accessible. Build MCP servers, invest in your API layer, make your product something agents can work with. Products that resist agent integration will lose to custom agent-built alternatives. Products that embrace being both a human workspace and an agent tool layer are the hardest to replicate.

Think about what your product looks like as a data provider, not a UI. If an agent bypasses your interface entirely and just queries your data, what’s left?

If the answer is “nothing,” you have a problem. If the answer is “our data relationships, compliance infrastructure, and collaborative workflows,” you’re probably fine.

Don’t build your own siloed agent. The interoperability problem kills this approach. Build connectors to the major agent platforms instead. Let the general-purpose orchestrators handle cross-tool coordination.

Budget for the transition, not just the technology. The hardest part of this shift isn’t technical. It’s organizational. How your team works, how your operations run, what you automate vs. what stays human: these are the questions that actually determine whether the transition works.

None of us know exactly which future plays out. But the companies that treat this as an architectural and organizational challenge, not just a feature roadmap item, are the ones I’d bet on.

I’ll be back next quarter with an updated take. If the evidence shifts, so will my position.

If you’re figuring out what AI actually changes for your business, that’s what we do at Seeko. We help companies build AI into their architecture and operations: strategy, systems, and automations that actually work in production.