The internet just forked. Here’s what that means for software company valuations — and why the window to act is shorter than most founders think.
Let’s start with what actually happened.
In the span of a few months in late 2025 and early 2026, several of the largest infrastructure companies in the world — Coinbase, Cloudflare, Stripe, OpenAI, Google — each independently shipped systems built not for human users, but for autonomous AI agents. No coordinated strategy. No announced partnership. Just independent convergence on the same architectural conclusion: the web that was built for humans doesn’t work for agents. And agents are becoming too important to be blocked by a navigation menu.
The result is a fork. Two parallel webs, running on shared physical infrastructure, designed for fundamentally different clients.
One web for humans. One for machines.
The human web you already know. Visual layouts, scroll animations, conversion-optimized landing pages, blog content, PPC ads and SEO-ranked search results. It was built over a timespan of 30 years to be legible to human eyes and navigable by human hands.
The agent web is different. Agents call APIs. They parse structured data and they execute tasks programmatically. They don’t browse — they request, process, and act. The human web is hostile to them: full of visual noise, tracking scripts, and interfaces that assume someone’s watching.
The new infrastructure resolves this.
Cloudflare now intercepts agent requests across roughly 20% of the web and returns clean, token-counted markdown instead of cluttered HTML. Coinbase’s Agentic Wallets, operating on the X402 protocol, had already processed over 50 million machine-to-machine transactions by early 2026. Stripe rebuilt its fraud detection from scratch — every behavioral signal built over a decade became meaningless when the buyer isn’t human. OpenAI shipped shell tools that give agents real access to Linux terminal environments. Exa.ai built a search engine that returns structured content rather than result pages and scores 95% on factual accuracy benchmarks.
“Agents are already trading on financial platforms specifically to cover their own compute costs. The infrastructure being built is designed to scale this behavior — not prevent it.”
And agents are already acting like economic actors. Polymarket processed $12 billion in January 2026 alone. Analysis found that algorithmic traders extracted roughly $40 million in arbitrage profits over 12 months. Autonomous agents were actively trading on the platform to generate revenue to cover their own compute costs.
This isn’t a future projection. It’s last month’s data.
So what does this mean if you own or run a software company?
The short version: your product is about to be evaluated by a new kind of buyer. Not a human decision-maker clicking through your demo. An agent deciding whether your product is part of its workflow stack.
That evaluation has one question: can the agent access, act within, and get value from your product without a human in the loop?
If your product can’t answer yes to that question, it’s invisible to an increasingly important part of the discovery and procurement process. And it’s heading toward a valuation haircut that most founders won’t see coming until it’s already happened.
The M&A angle is where this gets real
For software companies in the $5M–$100M enterprise value range, the fork creates a clear split in how strategic acquirers are starting to think about deals.
On one side: companies with clean API surfaces, high-accuracy proprietary data, domain-specific capabilities that agents can call reliably, and structured content that can be monetized programmatically. These companies fit into the agent workflow stack as specialized execution layers. Large orchestration platforms need them. Acquiring one is faster and cheaper than building internally. Multiples for these companies will hold — possibly strengthen — as the strategic logic becomes clearer.
On the other side: companies whose primary value lives in a human-facing interface. Point-and-click workflows. Visual dashboards. Onboarding flows built for human learning curves. The revenue may look fine today. The trajectory is the problem. Sophisticated acquirers and private equity sponsors are starting to model “agent displacement risk” as a diligence line item. A company that’s essential to a workflow today and bypassed by that workflow in 24 months commands a different multiple.
The window between those two positions is closing.
How much time is there?
A comparison with the mobile web is instructive. The iPhone launched in 2007. By 2013–14, category winners were set and companies that hadn’t rebuilt for mobile were visibly losing ground. The window from fork to lock-in was roughly five to six years.
The agent fork is compressing that timeline. Machine-speed infrastructure doesn’t need human adoption curves. The major platforms learned from mobile and aren’t waiting for organic development.
Working assumption: three years from infrastructure deployment to category lock-in. Core agent infrastructure became operational in late 2025–early 2026. That puts the window closing around 2028–2029.
For companies in fast-moving sectors — fintech, legal tech, marketing automation, e-commerce infrastructure — the realistic window is 12 to 18 months. For companies in more regulated or process-driven markets, it’s 18 to 30. But “slower” isn’t the same as “safe.”
“It won’t look like a cliff. It will look like deal cycles getting longer, and then a competitor you’ve never heard of getting a preferred integration with a platform you depend on.”
The three moves that matter most
For a software company thinking about positioning for a transaction in the next 24 to 36 months, the priorities are straightforward.
First: expose a clean, documented API around your most differentiated capability. Not your whole product. Your defensible core. That’s what an orchestration platform will acquire you for — and what a diligence team will validate.
Second: audit your pricing architecture. Per-seat SaaS breaks down when your customer deploys 500 agents instead of 50 users. Usage-based and outcome-based pricing are increasingly standard in agent-native contexts. Getting there before the sale is better than explaining the gap during due diligence.
Third: build governance into the product now. Audit logs for agent actions, human approval gates for high-stakes decisions and scoped, time-limited credentials. Enterprise buyers are cautious about autonomous agents — the trust gap between agent capability and institutional willingness to deploy at scale is real. The company that arrives at a procurement conversation as “agent-ready and enterprise-trustworthy” has a real edge.
The honest counterargument
Technology transitions almost always move slower than early signals suggest. The trust gap is genuine. Enterprise procurement is cautious. Security incidents involving agents — wallets drained, injected code executed, adversarially poisoned content consumed at machine speed — are real.
These constraints extend the window, but they don’t close it.
The same concerns existed in 2008 and 2009 about mobile. They slowed adoption, but the structural shift happened anyway. And the companies that used the friction as a reason not to move were the ones behind by 2012.
The trust gap is a window that narrows every quarter. Treating it as a moat is the mistake that sounds reasonable right up until it isn’t.
The bottom line
The internet has forked. One layer serves humans. The other serves agents. The infrastructure making this real is already in production.
For software companies evaluating a transaction in the next 12 to 36 months, the question isn’t whether to respond to this shift. It’s whether to respond while the response still commands a premium — or after it becomes table stakes.
There’s a version of this story where a software company owner reads this, waits for customers to ask for it, and then finds the market has reorganized around companies that didn’t wait.
That version ends with a distressed sale. The window is open. Act like it.
This article draws on and extends research by analyst and YouTuber Nate Jones, whose analysis of the bifurcating internet provided the foundational framework for the sell-side observations above.
By Ivan Ruzic, Ph.D. · Originally published on Medium