The most revealing word in Untapped Ventures’ pitch is not “artificial.” It is “work.” The Los Angeles firm is trying to separate AI that decorates existing software from systems that take responsibility for a result. A chatbot drafts. An autonomous system decides what needs doing, runs the sequence, checks the work and delivers the outcome. That distinction is the organizing idea behind a $27 million first fund, 35 portfolio companies and initial checks advertised at $500,000 to $2 million.
This is a venture firm, not a product company, so its product is partly judgment: a concentrated way for limited partners to reach young AI companies, and for founders to gain capital before their category is settled. Founder and general partner George Bandarian brings the habits of an operator. Before venture, he ran B2B enterprise-software company AMI, scaled it and sold it in an all-cash transaction in 2018. The biography matters because Untapped’s support is built around practical chores - sharpening a category, testing a go-to-market motion, preparing a fundraise and finding an enterprise door that might open.
The filterAssistance is common. Accountability is rarer.
Untapped calls its market the Autonomous Economy. In the firm’s formulation, digital agents and physical robots do economic work end to end. A legal system would not merely summarize a contract; it would identify unusual clauses, suggest changes and conduct the exchange before a person reviews the result. An insurance product would not give a broker a nicer dashboard; it would perform the brokerage workflow. The firm is looking for a transfer of execution, not a thin AI feature pasted onto a familiar subscription.
That gives a sprawling portfolio a usable spine. The opportunity set includes agent identity and permissions, orchestration, observability and audit trails - the control layer that lets a company know which machine acted, under whose authority and with what result. It also includes outcome-delivering applications, robotics, industrial systems and the compute underneath them. Untapped’s public request for startups ranges across 12 frontiers, from “SaaS killers” and agentic infrastructure to life sciences, materials, defense, neurotechnology and energy. The sectors move; the execution test stays put.
The loop is the differentiator: capital travels out, market knowledge and introductions travel back.
The customer loopA cap table that can answer the phone
The firm’s more distinctive asset may be its limited-partner community. At the end of 2025, Untapped counted 99 LPs; its current language rounds that into a network of more than 100 enterprise CEOs. These are not presented as silent names on a quarterly letter. They help source companies, examine deals, advise founders, make introductions and, when the fit is real, become buyers.
That creates a compact flywheel. An executive who knows industrial procurement can spot whether an AI workflow solves the expensive part of the job or merely makes a good demo. A founder can hear objections before a formal sales cycle begins. The fund gets sharper diligence; the executive gets an early view of technology; the startup gets access to a person who understands the budget and the bureaucracy. It is customer discovery embedded in the capital base.
“Right after their first investment, Untapped connected us with their LPs that are actually industrial companies.”Philipp Wehn, CEO of portfolio company Nexxa.ai
The arrangement is useful precisely because early enterprise AI is awkward. Buyers worry about reliability, security, permissions, integration and who carries the blame when an agent does the wrong thing. A friendly introduction does not erase those questions. It gets the right questions asked sooner. Untapped’s founder support - direct strategic sprints, narrative work, fundraising preparation and distribution help - is meant to convert that early contact into a repeatable company rather than a pile of bespoke pilots.
The portfolioProof arrives company by company
A thesis this large risks becoming a costume that fits every deal. Untapped’s named examples make it more concrete. You.com moved deeper into enterprise AI and closed a Series C. Lemurian Labs raised a Series A to develop edge AI processors for robotics, industrial automation and critical infrastructure. Extropic released a chip based on thermodynamic computing, an attempt to change the economics of inference rather than squeeze another optimization from conventional architecture. Harper applies the end-to-end idea to commercial insurance. Nexxa.ai brings AI-native consulting to industrial companies.
The customers therefore vary. Some portfolio companies sell directly to enterprises that want a completed workflow. Others provide the chips, control systems and developer infrastructure those applications need. Untapped itself has two customer groups: technical founders seeking a first institutional partner, and LPs seeking private-market exposure to early AI. The model is classic venture capital - pool LP money, buy minority stakes and seek returns through later financings or exits - with fee and carry terms that remain private.
What a founder can steal
Pitch the job, not the feature. Name the workflow, show how much of it runs without intervention, identify the person who owns the budget, and explain what must be true for that person to trust the machine. The model matters; the handoff of responsibility matters more.
The competitive mapA specialist in a market full of specialists
Founders have many alternatives: broad pre-seed funds, AI specialists, accelerators, corporate venture arms and large multistage firms. Gradient Ventures can offer the gravity of Google. NFX brings a large early-stage network. Y Combinator supplies a recognizable batch and fundraising machine. Firms such as 2048 Ventures also pursue technically ambitious pre-seed companies. Untapped cannot win by being the loudest or richest. Its case rests on specificity, speed, hands-on operator work and whether the CEO network produces useful collisions.
For a founder, the practical fit is reasonably clear. Untapped is most relevant before a market has an agreed name, when the technical insight is stronger than the sales machinery and a $500,000-to-$2 million check can still shape the round. The founder must want an involved investor and be prepared to expose the product to operators who know the workflow. A team seeking a passive cap-table name, a late-stage balance sheet or a consumer-growth playbook may find a better match elsewhere. The firm openly favors technical depth, rare domain credibility, speed and what it calls an “earned secret” about a market.
There are tensions. “Autonomous” is an appealing label in a market that routinely renames ordinary automation. End-to-end systems may also require more service, compliance work and human supervision than the clean thesis suggests. Selling outcomes can improve alignment, but it can also transfer operational risk from customer to vendor. And a portfolio spread across software, robotics, chips, biology and energy demands very different technical judgment. The thesis earns credibility only when those companies reach dependable deployment.
Investors face their own version of the choice. Untapped offers concentrated access to a young and volatile category, not the diversification of a broad technology index. Its community model asks more from LPs than wiring capital and reading updates; the useful member is expected to share pattern recognition, evaluate a market or take a founder meeting. That participation can improve information flow, though it does not remove the long holding periods, illiquidity and high failure rate inherent in seed investing.
Untapped seems aware that conviction can calcify. Before publishing its 2026 Autonomous Economy thesis, Bandarian wrote that the firm stopped new deals, term sheets and partner meetings for a month of research. The resulting shift from “AI-native” to “AE-native” was a narrowing device: invest where software or machines execute, not everywhere AI appears. A pause is not proof of foresight. It is, however, a useful institutional habit in a business paid to act certain before the evidence is complete.
Where it fitsThe small fund at the workflow handoff
Untapped sits between a micro-fund and a broad seed platform: large enough to write a meaningful first check, small enough to make founder access part of the offer. Its Los Angeles base gives it proximity to aerospace, entertainment, logistics and a growing technical scene, while its dinners and events extend into San Francisco. The firm reported six new investments and 13 follow-ons in 2025, suggesting that reserves and existing relationships matter alongside new logos.
Its market bet can be reduced to a clean economic observation. Businesses do not ultimately want software seats; they want invoices reconciled, candidates screened, claims processed, equipment maintained and decisions made. For decades, software improved the humans doing those jobs. Untapped is financing the moment when a vendor offers to own more of the job itself. If that happens, revenue may migrate from software budgets and service labor into a new class of outcome businesses. The winners will need more than capable models. They will need trust, controls, distribution and the nerve to accept responsibility.
That is why the LP network and the infrastructure thesis belong together. Autonomous products need demanding early customers, and demanding customers need evidence that autonomy can be governed. Untapped’s experiment is to put both sides in the same room early, then supply capital while the argument is still productive. The result is less a futuristic robot parade than a familiar enterprise scene: a founder, a buyer and a difficult workflow, now negotiating exactly how much work the machine can be trusted to finish.