11.18.2026

Meet the Nine Companies Building for the New Agentic Economy

Jordan Crook

Today, nine companies have presented at our 15th Betaworks Demo Day. 

From when we launched our accelerator program in 2016 to today, there has been a proliferation of various other accelerator programs. But one thing that has remained unique to the Betaworks program is our thematic approach. 

We make it a point to orient around the fast-evolving foundational technology that is unlocking new capabilities at the application layer. 

Back in January of 2024, we had our first ever Agent-focused cohort. At the time, it was very focused on the infrastructure required to give LLMs the ability to act. Think headless browsers, agent observability, etc. 

From there, we moved more towards the application layer of agentic AI, investing in cohorts of vertical AI agents and Agent Systems, the earliest iterations of what we now think of as agent swarms. 

Today, we're focused on what we call the New Agentic Economy, which encompasses physical AI, trust and verification layers, UI, guardrails, commerce and more. 

The following companies have gone deep on these problems. So, without further ado: 

Ophelia

www.opheliaos.com

Infrastructure enabling agentic transactions, from intent to purchase

Problem

As software becomes increasingly capable of understanding what people want, the bottleneck shifts from intelligence to action. But there is no universal infrastructure for agents to reliably act on human intent across the internet.

Solution

Ophelia is the infrastructure enabling agents to turn intent into action handling authorization, navigation, transactions, and reliability required to actually complete user requests. Our API offers developers an easier way to access and build capabilities into their products.

Team

Intrinsic Labs

intrinsic-labs.ai

Intrinsic Labs turns an industrial company's operating knowledge into AI workers that do the work.

Problem

Every industrial business runs on an invisible operating system. Their knowledge is spread across people, inboxes, documents and disconnected software. That knowledge keeps the business moving, but it's hard to scale and nearly impossible for AI to use. Even companies that want AI agents lack the talent to build, maintain and run them.

Solution

Intrinsic Labs builds the infrastructure to turn that operating knowledge into AI workers that operate within a company's existing systems and workflows. Our model- and interface-agnostic platform gives those workers the context and controls they need to perform useful work alongside people. Our forward-deployed engineers supply the talent industrial companies don't have, and we deploy, run and improve the workers for them.

Team

X1 Wealth

x1wealth.com

X1 is an AI family office for families whose finances have become a second job.

Problem

Once a family has a business, a few properties, a trust and private investments, no one is holding the whole picture. The CPA, the attorney and the advisor each see one piece, so someone in the family ends up tracking the capital calls, the K-1s and the deadlines nobody else is watching. The wealthiest families hire a family office for this; most families with this much complexity never will.

Solution

X1 is that family office, built as software. It brings a family's accounts, documents and entities into one record the family owns, prepares the work coming due, and acts only when the family says yes, down to preparing a bank transfer for them to approve. It remembers what the family decided and why, and works alongside their CPA, attorney and advisor, and AI assistants like Claude and ChatGPT.

Team

Blobfish

blobfish.ai

Training Gyms for AI Agents

Problem

Imagine hiring someone who speaks a hundred languages, writes software, and understands quantum physics. And giving them one job: Check invoices. We are building training gyms for creating specialized model that’s cheaper, more accurate and can run on-prem.

Solution

Blobfish turns licensed software, company data, and records of agent failures into world most boring video game where model can practice fake enterprise task on a simulated company. Think of a flight simulator for an AI employee. For an accounting task, we recreate the documents and tools. The agent investigates a discrepancy, takes actions. If they get it right, we reward them with candy, if they got it wrong we whip them. We have leaderboard for the models and as they clear the level, we generate new level and adjust the difficulties.

Team

Heard

www.heard.dev

Building the future of human-computer interface

Problem

When we each work with 100 agents in the future, we need a voice that comes with context, knows what matters to you, and what doesn't.

Solution

Building the next voice interface for a billion users.

Team

Yaven

www.yaven.ai

AI teammates for companies of one.

Problem

The white collar job as we know it is dying. In ten years most people won't have a boss. They'll have clients. Millions of people are already going independent to do the work they love, and then they end up spending half their week on admin, chasing invoices and hunting for the next client. AI was supposed to fix this. Instead it handed everyone a blank box and said "ask me anything," and most people have no idea what to ask.

Solution

Yaven is the team every independent never had. It works out what you're actually trying to do before you have to ask, and shows up right where you're working, in your Mac's notch or one key away from any app. It finds your next client, runs everything after the yes, and learns your clients, your voice and your taste, until one person can run what used to take twenty. You get your hours back for the work you started this for.

Team

Embarcadero Labs

embarcaderolabs.com

Embarcadero is building agentic physical AI that supports frontline workers on every job today and harnesses what it learns to enable the future generation of trades.

Problem

AI is fast and accurate on a screen, but on a real job it hits a wall: it either answers in one shot and gets it wrong, or re-plans so often it’s too slow to help. What it’s missing is the knowledge that lives in the physical world, like what’s on site, what was done last time and what the customer is owed, which today sits in workers’ heads and in paperwork written after the fact. So field service companies can’t scale with AI: technicians lose hours to reports, work orders and contract obligations slip, and quotes and delivery depend on whoever is on the job.

Solution

We’re building a multimodal AI assistant that works through a technician’s phone, Bluetooth mic or smart glasses while they do the job. Its core is a prioritization engine: it weighs video, voice and photos from the job against the customer’s history, keeps the evidence that matters, flags what’s missing, and fills in the company’s own reports and forms with each answer linked back to the moment it came from. Every job adds to a knowledge graph of customers, sites and equipment, so the next worker, starts with everything the last one learned.

Team

Pakket

www.pakket.ai

Warehouse Automation using Large Vision Models

Problem

Most warehouses only run at 65% efficiency because they do not have complete visibility into site operations. This leads to errors, edge cases and mistakes.

Solution

Large Vision Models to improve warehouse efficiency and reduce errors

Team

Likert

Stealth

Team: Arda Tasci

‍

Thank you to our Camp sponsors, Mercury and Carta.

REad more

Back to writings