Muse, Myself and AI

On September 29th, Meta launched Muse for Small Business. It plugs into Shopify, Dropbox, Slack, QuickBooks, Stripe, Klaviyo, and Notion, and connects directly to Instagram professional analytics, Facebook Pages, and Meta ad accounts.¹ The launch came a day after Meta unveiled its Meta Enterprise Platform, and it follows the consumer version of Muse, which hit No. 1 among free iPhone apps in the US within ten days of release.²

Two years ago, we participated in Dev/Agents’ (now Dreamer) seed round, the team and infrastructure now powering Muse. Our thesis at the time was that trust would consolidate the consumer AI market into a small number of dominant platforms because value comes from full access to your email, calendar, contacts, browsing history and financial information. That thesis still stands, but Muse for SMB points to a new boundary in software.

The defining distinction is no longer consumer versus enterprise, but single-principal versus multi-principal businesses.

The Enterprise Gold Rush

For the past decade, venture has been walking away from consumer and toward B2B for one simple reason: businesses pay and consumers mostly don’t. Using Y Combinator’s own industry tags, B2B companies went from roughly 40% of the 2015 batches to 65% in 2025, while consumer fell from about 22% to at most 8.4%.³

The AI labs are following the same money. Anthropic has chased the enterprise from day one, and doubled down this year by extending the agentic approach behind Claude Code to the rest of the office with Claude Cowork and plugins for finance, legal and HR. OpenAI has spent 2026 catching up, pushing ChatGPT from consumer household name into the enterprise. It launched Frontier, its enterprise agent platform, then ChatGPT Work, and most recently Dots, always-on agents with their own cloud computers, for Pro and Business Premium users.

As we laid out in Forward Deployed Lock-In, the fact that both labs are sending forward deployed engineers into midmarket companies is an admission that raw capability alone isn’t enough. To become the enterprise horizontal operating layer, they need to understand the deeply entrenched data schemas, historical context, and user behavior patterns that businesses have built up over decades. So, while OpenAI and Anthropic pay engineers to learn how businesses work, Meta is betting millions of small business owners will teach it for free.

All the Budget, None of the Permission

The labs have found the buyer with budget, but what they haven’t solved for is delegation. Inside an enterprise, nobody has a clean answer to who pays when the agent gets it wrong. The employee who connects the agent is almost never the person who owns the customer relationships, the data or the contracts that are now at risk. That is why enterprise contracts take quarters to close, and why deployments arrive wrapped in forward deployed engineers, audit trails and long, staged rollouts.

Enterprises handle this the way they handle any other risk, with least-privilege access. That contains the downside, but it also caps the upside. An agent that can only see a sliver of the business can only make decisions about a sliver of it.

Consumers, meanwhile, will hand an agent the keys to their whole digital life without reading a word of the terms. Meta’s personal agent Muse reached 5 million downloads in 22 days, outpacing ChatGPT, Grok and Claude, which took 56, 103 and 492 days respectively to reach 5 million U.S. downloads.⁴

The reason is simple. When the data owner, the decision maker and the liable party are the same person, delegation takes a single tap, and the system receives total operational context.

The upside of handing over your email, messages, location and screen is a tireless assistant. One user of Instinct, the $10 billion personal agent startup, reported that it watched a sold-out museum for three days, bought tickets the moment spots opened and rearranged their dinner plans around them.,⁵⁶ The downside is that the user is solely responsible for whatever the agent does on their behalf, whether that’s the $200 cancellation fee on a restaurant Instinct booked or Muse accepting a lowball Marketplace offer and sharing a seller’s home address.^{7 8}

The catch is that consumers don’t pay, and Sora’s shutdown was a reminder that topping the App Store isn’t a business model.⁹

Meta’s Shortcut to the Enterprise

Enterprise has the revenue but cannot delegate. Consumer delegates but struggles to monetize. Meta’s Muse for SMB targets the one segment where business budget coexists with consumer-level decision speed: the single-principal firm.

The labs send forward-deployed engineers (FDEs) into organizations with many principals to map how the business runs, build the workflow software, and push token volume. Meta hands that same job to millions of independent business owners. They teach the model just by running their businesses, and they absorb the cost when it gets something wrong.

This gives Meta four advantages:

1. Consumer liability profile: With one principal there’s no procurement process, no legal review and no buying committee. The owner taps accept and the agent is working that afternoon.
2. Context on day one. Meta already has the Instagram analytics, pages and ad accounts, and trains on real workflows by default. An enterprise agent has to earn that kind of access one permission at a time.
3. The labs take the expensive fight. PE roll-ups suit FDEs because authority is fragmented, but rebuilding a single decision maker on site is slow and costly. SMBs already have a single decision owner.
4. No subscription reliance: ~98% of revenue comes from ads, spread across 10M+ advertisers, so Muse for SMB can stay free and earn through ad spend.

The single-principal business is the easier adoption bet which is exactly why Meta went there first. But a market that’s easy to enter is also easy to crowd, and Meta has just reset the price of the assistant to zero. If adoption takes one tap, so does switching. Any startup charging an SMB a monthly fee for “an agent that runs your business” is now competing with free, plus better distribution and day-one context.

Muse-Proofing Your Startup

Instinct processes roughly $1B in annualized transaction volume, 40% of its users have connected a credit card, and the average monthly spend is $1,300.,¹⁰¹¹

This is proof that consumers will spend. They just lack the appetite for another subscription.

Meta entered the single-principal market with its consumer playbook: give the agent away, subsidize it with ads, and use distribution to win the surface layer. But free only beats paid when it does the same job and Muse still stops short of the operational complexity that businesses will pay to solve. The challengers worth watching are going after exactly that gap.

1. Meta earns when the owner spends. Challengers earn when the owner earns. Muse is built around what Meta already does well: content, discovery and ads. The messier work of actually running a small business (taking orders, moving money, answering the phone, keeping customer data) sits outside Meta’s core and isn’t something it’s set up to solve.

Avoca picks up where the ad spend ends for HVAC, plumbing and roofing companies, answering the leads those ads create and showing owners which ones turn into booked jobs and paid invoices, so they know exactly where their next marketing dollar should go.¹²Today, while Muse is limited to outbound calls, Beside reached $4 million in annual recurring revenue with more than 20,000 paying customers by handling inbound calls for realtors, contractors, salons, insurance agents and other service businesses.¹³ Arini is doing the same for dentists, turning calls that used to go to voicemail into $39K a month in production for Normandy Lake Dentistry.¹⁴Owner (Activant portfolio company) runs an independent restaurant’s entire digital storefront as a single system, from the website and online ordering to the branded app, loyalty program and point of sale. It helps to keep a restaurant’s costs low with a month-to-month subscription and earns its upside from a 5% fee paid by diners on each order, which is a third of the up to 15% that delivery apps charge.¹⁵ Owner only grows when the restaurant’s direct orders do.

2. Meta caps its liability. Challengers cap the owner’s. Muse’s terms make the owner solely responsible and cap Meta’s exposure at $250. That’s fine for booking museum tickets but doesn’t work for an agent buying inventory. Naive gives each agent its own card with a hard spend limit, so the worst case is set by design, not by a terms of service page. An owner who knows the maximum loss delegates more, and Naive earns on that spend.

3. Meta sees the storefront. Challengers own the books. Muse reads Instagram analytics, pages and ad accounts. That’s the marketing layer, not where the high-stakes decisions sit. Kick replaces QuickBooks rather than plugging into it, so its agent works from the ledger itself. Meta sees where the owner’s attention goes. Kick sees where the money goes.

4. Meta arrives at launch. Challengers arrive at birth. Meta’s day-one context assumes an Instagram page and an ad account. A business started this week has neither. Result (YC) manages incorporation, accounts, payments, product launch and day-to-day operations in one place, with agents doing the work alongside the founder. While the better known Polsia ‘runs your company while you sleep’, charging $50/month plus 20% of transactions through the business, with ad spend excluded. Both own the record from the first dollar.

The Principal Difference

Single- versus multi-principal maps the agent market far better than consumer versus enterprise. A sole proprietor running a $5 million business can delegate more like a consumer than an employee at a 50-person company because the owner controls the data, makes the decisions and absorbs the downside.

Meta isn’t moving upmarket so much as extending its consumer playbook to the businesses that still behave like consumers. In that single-principal market, the intelligent assistant is heading toward free. The startups that can still charge will be those that go deeper into the operational work Muse won’t own, where money moves, records live and risk can be bounded. Muse found the B2B agent market’s shortcut: one principal, one tap, and permission to act. Everywhere else, adoption waits for liability, controls and insurance to catch up.

Footnotes

1. Meta, The Future Is for Everyone: Muse for Small Business, 2026 ↩

2. TechCrunch, Meta’s Muse is outpacing ChatGPT’s early mobile launch, 2026 ↩

3. Y Combinator company directory, industry tags by batch, Activant analysis. ↩

4. Forbes, Meta’s Muse AI Assistant Hits 5 Million Downloads—Outpacing Growth Of ChatGPT, Grok And Claude, 2026 ↩

5. Verdict, AI startup Instinct raises $1bn in Series C at $10bn valuation, 2026 ↩

6. The Atlantic, It’s All Fun and Games Until You Give AI Your Credit Card, 2026 ↩

7. Ibid ↩

8. The Guardian, Meta’s AI agent Muse gives out user’s home address without permission, sending buyer to his house, 2026 ↩

9. OpenAI, What to know about the Sora discontinuation, 2026 ↩

10. BigGo Finance, Noah Shinn: The $1 Trillion Agent Economy Will Make Compute the New Oil, 2026 ↩

11. X, Noah Shinn: Instinct + Stripe, 2026 ↩

12. Avoca, Why We Are Building, 2026 ↩

13. Beside, Beside Raises $32 Million to Turn Trillions of Calls and Texts into Growth for Everyday Businesses, 2026 ↩

14. Arini, How Normandy Lake Dentistry Turned Missed Calls into $500K in Production, 2026 ↩

15. Owner, Simple pricing that fits your restaurant, 2026 ↩