We asked founders which seed-stage startups they’re watching. Here are their picks.

There is an old maxim in the culinary world: "If you want to know where to eat, ask a chef."

Felicis (opens in new tab) and Newcomer (opens in new tab) decided to apply the same adage to early-stage companies. If you want to know which seed-stage startups are poised for success, ask other founders.

So we did. Emergent (opens in new tab) is a list of seed-stage startups nominated by founders, compiled by Newcomer and Felicis. The results yielded a clear, but nuanced, take on the future of AI.

The list reveals a thriving seed-stage ecosystem, even as industry focus has increasingly shifted toward larger, later rounds. Global seed funding totaled $13 billion in Q3 2026, Crunchbase data shows (opens in new tab). Of that, $2.6 billion went to seed rounds of $100 million and over, around 20% of seed financings. This last quarter’s totals are up 44.4% compared to $9 billion in seed funding (opens in new tab) in Q3 2025. (Felicis is an investor in Crunchbase. It has no say in its editorial process.)

The founders’ seed-stage nominations span AI tools that help people write software, create music, and work with better information. They also show enthusiasm for bringing AI into the physical world, with picks ranging from robots designed to work in factories and space to companies applying AI to biological research and energy infrastructure.

Unsurprisingly, AI infrastructure-related startups led the nominations; robotics and physical AI were also prominent, while biology and healthcare stood out as a welcome surprise among startups to watch.

Building the infrastructure for a world outside of the chatbox

The AI infrastructure companies spanned a variety of subcategories, including payments, security, monitoring, model routing, inference, data and context tools. Felicis’s Feyza Haskaraman (opens in new tab) isn't surprised the category dominated the list.

“Agents have left the chatbox and are now doing longer, more important work. The magical experiences, from personal assistants to enterprise agents, depend on having the right foundations in place,” she said. “It’s become very clear that the companies that nail memory, routing, and security are the ones winning.”

The picks highlighted a clear focus on memory, which Feyza believes makes sense.

“Memory is where agents fail most visibly, and also where they delight most visibly,” she said. “Get it right, and you cut costs too, because the best memory uses fewer tokens.”

Overall, the AI infrastructure nominations reflect a shift in how people find and use products, with agents increasingly acting on their behalf.

In nominating Armature (opens in new tab), Vercel (opens in new tab) founder and CEO Guillermo Rauch described a world in which users reach products through Claude, ChatGPT, and coding agents.

“Armature captures every session on your MCP server, Claude Connector, or ChatGPT App, and shows what users ask, what agents think, and how they use product,” he said.

Useful agents need context – and a way to earn trust

Across the categories, three themes recurred: the need to supply context, establish trust, and adapt AI into the particulars of a job.

“We believe context and knowledge – fresh context and real, grounded truth – are extremely important for anything to be useful,” said Felicis's Peter Deng (opens in new tab).

Models need access to current information and relevant knowledge about their users to do useful work, Peter said. For specialized applications, he believes that understanding the customer and the workflow can be just as crucial as the underlying model’s capabilities.

“If you’re building a company in a vertical that’s specialized,” noted Peter, “you really have to make sure: Are you fitting the workflow? Do you know the customer well enough? Are you actually getting to the root of their problem?”

Several nominations echo that viewpoint. Imbue (opens in new tab) co-founder Kanjun Qiu picked Tako (opens in new tab), which organizes data so language models can use it.

“They've aggregated a ton of data across many different sources and cleaned it up so LLMs can actually read it in sensible ways,” Qiu wrote. “You can ask pretty insane analytics questions to Tako and they answer with quality data.”

Snorkel AI (opens in new tab) co-founder Alex Ratner nominated Attention Engineering (opens in new tab), which captures context from people’s computer activity to inform more personalized AI.

Other founders focused on cost and control in a world where multi-model usage is the norm (opens in new tab). Mercor co-founder Brendan Foody (opens in new tab) nominated Martian (opens in new tab), whose model router selects a model for each request. ElevenLabs (opens in new tab) co-founder Mati Staniszewski chose Arrakis Security (opens in new tab), which monitors agents inside enterprise systems and allows companies to intervene in their actions.

“Security tooling was built for humans using machines; agents now act autonomously inside CRMs and financial systems,” he wrote.

Reid Hoffman’s pick, Mnemom (opens in new tab), records agents’ actions and checks them against company instructions.

“As agents do more of the world's work, trust in them will have to be earned from a record, the way it is with people,” he wrote.

Growing interest in robotics puts scale in focus

Some of the nominations reflect that shift toward AI performing physical tasks. Tines founder Eoin Hinchy (opens in new tab) nominated Icarus Robotics (opens in new tab), which is developing AI-powered robots for work in space, a setting “where human labour is incredibly expensive and constrained.” Reid Hoffman also selected MicroAGI (opens in new tab), which develops intelligence for manufacturing robots.

“I've been interested in how AI is moving beyond software into robotics and the physical world,” Eoin wrote in his nomination of Icarus. “It's an ambitious application of AI.”

Felicis’s James Detweiler (opens in new tab) poses a thought experiment: what if we’re nearing the end of a chapter in human history defined by knowledge work and days spent behind a computer?

“Could we see a higher proportion of jobs in the ‘real world’ in the future? Might this warrant more ‘real world’ investing today? I think so,” he said. “But one nuance to remember: digital AI accelerates physical innovation. A quantum computing CEO told me AI lets his 20-person R&D team do the work of 50. In a field with only dozens of experts, that’s a really big deal. Expect faster real-world breakthroughs in the future.”

Nominations were rooted in distinct settings for physical AI, from space to the factory floor. Beyond those individual applications lies a broader question: As these robotics companies begin to scale, how can they spread the cost of developing their technology across enough machines to make the economics work? James sees an opportunity in models that can learn from – and improve – many robots at once.

“Suppose one robot brain runs across many bodies. You train a model once, every deployment generates more data, and improvements can propagate across the entire fleet,” he said.

Investor interest in robotics has surged. Global robotics startups raised roughly $31 billion in the first nine months of 2026, compared with $10.9 billion in the same period of 2025, an increase of about 185%, according to Crunchbase data.

But turning that investment into dependable machines is a long game. James cautions that investors new to robotics can underestimate supply-chain, financing, and commercial timeline risks. He expects the physical AI boom to unfold over decades.

Biology and healthcare AI move from models to experiments and care

While experts and pundits have been excited about the life-saving opportunities that AI unlocks for biology and healthcare, investment has historically been slow to follow. Until now.

Biology offers another arena to see whether AI can produce useful results.

“There are massive public benefits associated with using AI for biological and healthcare advancements, not seen in other parts of the market,” noted Felicis’s Tobi Coker (opens in new tab). “This heavily incentivizes foundation model companies to invest in these domains.”

Investors clearly see the potential. Investment in AI-focused healthcare and biotech has surged in 2026. By October 30, 2026, companies in those categories had raised $20.2 billion globally, already exceeding the $15.8 billion invested during all of 2025, per Crunchbase data. Biotech IPOs have also seen a strong resurgence with 22 companies raising $6.9 billion through September, according to a recent PwC report (opens in new tab).

Fireworks AI (opens in new tab)co-founder Lin Qiao nominated Phylo (opens in new tab), which adapts open-weight models to scientists’ needs.

“By customizing open-weight models for biology use cases, they give scientists the control and expertise they need at a fraction of the cost,” Qiao wrote.

David Luan selected Instance Bio (opens in new tab), whose technology tests many interactions between candidate protein binders and targets in parallel.

“We need to close the loop on bio models’ ability to run real-world experiments,” David wrote in his nomination.

Interest extends to clinical work as well. Glean (opens in new tab) CEO Arvind Jain nominated Almanac Health (opens in new tab), which grounds AI in reliable medical evidence.

“Phylo represents horizontal reasoning in bio for the first time ever; Instance represents data structuring and processing at scale; Almanac represents how AI can accelerate point-of-care,” Tobi reflected on the nominees. “The biggest opportunities still exist in the workflows where data is currently fragmented or judgment is still relatively manual."

Creator and workplace tools succeed by understanding how people work

For AI creator tools, usefulness depends on how well they fit how artists work. Runway CEO Cristóbal Valenzuela (opens in new tab) picked Resin (opens in new tab), which generates musical parts that fit an artist’s existing track, and Cartwheel (opens in new tab), which produces character animation for established studio pipelines. Both tackle specific steps in the creative process.

“It's much closer to a creative partner than a pure music generator, and I see a lot of similarities to how we've approached creative tools at Runway,” Valenzuela wrote of Resin.

Mati Staniszewski’s pick of Wonder Studios (opens in new tab), an AI film studio, extends that focus across an entire production. The company combines AI with live-action filmmaking and animation to produce films, music videos and advertising campaigns, working with filmmakers and artists to take projects from concept to finished production.

Peter noted that network effects play a crucial role in the growth of creator tools.

“We’ve seen this in the past wave of technology. Instagram gets better for every new creator who joins the platform. The whole thing gets better,” he said. “Network effects are where every additional user or every additional piece of data makes the entire product better. That is the compounding value.”

Beyond creative work, Reid’s nomination of SkillBench (opens in new tab) addresses a related question: How can companies adopt AI while continuing to develop their employees’ skills? The startup is building software that analyzes how developers work with AI coding assistants, tracking human and AI contributions to help employers understand where the tools improve productivity and where employees’ expertise may be weakening.

“The biggest AI returns will go to companies that also expand their employees’ individual potential,” Reid wrote.