There is a version of the future that is already being built. You just have to know where to look.
One of the underappreciated advantages of investing at pre-seed is the view it gives you. By the time a trend hits the mainstream venture conversation - the Series A/Bs, the thinkpieces, the conference panels - we've usually been staring at it for 12 to 18 months already. The conversations we have with founders today are a remarkably accurate preview of where capital, attention, and infrastructure are heading tomorrow.
We started investing in AI applications in early 2023 as investors raced to respond to the GPT moment. The smartest technical founders and early investors in the world are voting with their time and careers on what they believe will matter next. When a team of ex-Stripe engineers leaves to build financial infrastructure for autonomous agents, or a founder who spent six years processing $1T in reconciliations starts a company to automate the work he used to do by hand, those are leading indicators of the types of companies that will be dominating the headlines in subsequent months.
Here is what we are seeing today, and why it should matter to anyone investing downstream.
The Machine Layer Is Generating a New Infrastructure Stack
Cameron wrote recently about what we are calling The Machine Layer of the Internet - a parallel web being built for machines rather than human eyes. Discovery, communication, payments, and identity are all being rebuilt for a new kind of participant: AI agents and automated systems that transact without a human in the loop.
That thesis came from meeting the founders who are building it.
In the last few months, we have met companies building at every layer of this emerging stack.
One of the early shoots in this trend was companies building infrastructure for autonomous transactions. We saw this in early 2025, well before the incumbents like Ramp launched agentic cards. Around that time, we wrote the biggest pre-seed check into Natural, which is building B2B payment rails purpose-built for agents, settling synchronously instead of routing through human invoicing workflows. When we first saw that company, there was no Visa Trusted Agent Protocol, no Mastercard Agent Pay. The founders were building for a world that the incumbents hadn't publicly acknowledged yet. Now there are new headlines daily about agentic payments and commerce, and companies like Stripe, Shopify, Visa and more are scrambling to define and build their strategy in this area.
One new trend we're seeing is agents moving past bespoke, highly verticalized, and insular workflows into much broader, generalist scope. Companies like Town, and Cofia, another Restive portfolio company, are enabling non-technical users to broadly integrate AI into their day-to-day workflows with no friction. A truly autonomous co-pilot for your daily work - one of AI's most elusive white whales - is much closer to reality than most people realize. Tools like these are what will be able to take root and unlock latent demand from workers across the world, not just in Silicon Valley circles.
AI governance is emerging as its own category of company as well. A whole slew of products is emerging around this theme, and we’ve already backed companies like Fairplay that are building here. As agents gain the ability to take real-world actions - sending payments, signing contracts, accessing sensitive data - the question of who controls what they're allowed to do, and what data goes into the decisions they output, becomes urgent. We’ve also seen founders build insurance products for AI risk – not using AI to price insurance – but pricing the actual risk of AI hallucination within regulated financial institutions. It’s becoming increasingly clear that building companies that manage AI, not just use it, is becoming increasingly frequent.
The Permission Layer Needs to be Built
Tyler wrote about the Permission Layer - the deterministic barrier that needs to exist between AI systems and real-world actions. LLMs are inherently stochastic. Any instruction given to them is "soft." You cannot reliably constrain an agent's behavior from within the agent itself. The permission layer must exist outside the model, and this is infrastructure that must be built.
This is showing up in our conversations with founders as a concrete product category. Founders are building companies around the premise that as agents get more capable, the control infrastructure around them becomes both the bottleneck and the opportunity.
Our portfolio companies are already building pieces of this stack. Crossmint is creating agent-native payment credentials. Flowglad is building programmable money designed for machine-driven commerce. Together, they represent the early scaffolding of a permission and transaction layer that will need to exist before agents can operate at scale in financial services.
As Tyler put it: without this layer, agents remain a niche product for techies and early adopters. The companies that solve granular, customizable, secure approval workflows for autonomous systems will be the ones that unlock mainstream adoption by removing the friction keeping agents siloed today.
What This Means for Downstream Investors
If you invest at Series A and beyond, the companies hitting your inbox soon are being formed right now around these themes: identity, discoverability, and payments for agents, AI native workflows, permission and control layers, general purpose agents, and other new financial primitives built for a machine-driven economy. The founders building in these spaces are often the same engineers who built the incumbent infrastructure.
The speed is striking. As Cameron noted, the entire machine layer stack - from discovery to communication to payments to identity - went from concept to production in roughly fourteen months. The API layer took a decade to mature. The companies that move first will establish positions that compound quickly, and the window to shape the architecture is narrow.
As pre-seed investors, we occupy a unique vantage point at the frontier of innovation. The companies building agentic infrastructure, generalist AI, and governance frameworks for autonomous systems are laying groundwork that will drive capital, attention, and talent into action sooner than most expect. What emerges after that is almost impossible to predict.
© 2026 Restive®, Inc.
