Hi Reader,
I wanted to share a realization that's been rattling around in my head all quarter: the best-performing companies in our fund have something in common that I never planned, never wrote into a thesis, and honestly would have bet against as a former founder. They're marketplaces (coincidentally the type of business of I tried to build a decade ago).
Below is my attempt to explain why the model that broke my generation of founders suddenly works, what AI actually changed, and why the result, against every doom headline, is more people hired, not fewer.
As always, we welcome your comments, thoughts, and feedback!
— Shruti
Marketplaces Were Broken for Twenty Years.
AI Fixed the Broken Pieces.
And the output is more people hired
Two things are true
Two things are true in our portfolio that aren't supposed to be true at the same time.
First: our fastest-growing companies are marketplaces, the business model my generation of founders was told to avoid, the one that famously devours operating budgets and dies waiting for network effects.
Second: the AI inside them isn't taking work away from people. It's handing work out: hospitality workers placed into shifts they couldn't access before, doulas matched with families who couldn't find them, contractors paid to modify homes for people aging in place, creators keeping more of what they earn and people connected to benefits dollars they were owed but couldn't navigate to.
These two facts are not a coincidence, and explaining why is the point of this piece.
I know, because a broken marketplace is where I started.
The hardest way to learn this lesson
Before Symphonic, I co-founded and built a used furniture marketplace. I know firsthand how challenging building a marketplace can be, and ours was no exception. We were managing physical products and dealing with inventory management, warehouse operations, and logistics, in addition to keeping both sides of the marketplace in balance. As a result, our operations team grew every time GMV did. So when I say the most interesting thing happening in early-stage software right now is happening to marketplaces, understand that I'm the last person who thought this would be true.
Companies that five years ago would have been built as static SaaS tools are being built or rebuilt as dynamic marketplaces. This is a structural consequence of what AI made possible
The two eras, in one paragraph
You already know the great marketplaces: Uber, DoorDash, Airbnb, Amazon. They don't sell you software (with the exception of Amazon, that does sell software), they run the market itself. Uber didn't sell dispatch software to taxi companies. Amazon didn't sell inventory tools to shops, it became the mall and eventually built a software platform to subsidize the cost of its marketplace. Notice what those wins cost: tens of billions raised, both sides of the market subsidized for years, and armies of operations people behind the apps. Running a market is operationally brutal, and only the most colossally funded companies survived long enough for the economics to turn. So everyone else got the same advice, advice I lived: don't build the marketplace, build the tool. Sell software to the businesses in the market and let them keep doing the hard part themselves. That's what a SaaS business does. It’s Shopify selling you the store rather than Amazon being the mall. The industry settled into an equilibrium: SaaS got the margins, marketplaces got the market, and almost nobody got both.
The tool next to the transaction
That equilibrium put a ceiling on vertical software that nobody liked to talk about. Software could organize work, but it couldn't do work. Take staffing: the classic play was B2B workforce software — scheduling, timekeeping, compliance, maybe a job board bolted on. You sold it to the restaurant for a few hundred dollars a month, but the actual problem, finding a qualified, reliable bartender for Saturday night, remained entirely the customer's problem. The software was a filing cabinet for a labor market it never participated in.
Why did founders stop at the tool? Because stepping into the transaction meant operations: screening thousands of workers, matching them to shifts, catching fraud, setting prices, handling the 2 a.m. no-show. That work took humans, lots of them. The ops team scales with volume, the software margins evaporate, and the network effects arrive years after the payroll bills do.
And we don't have to speculate about how that movie ends, because a whole cohort ran the experiment in the 2010s. Shiftgig raised $56 million to build an on-demand staffing marketplace, grew past two hundred employees, then retreated, selling off its staffing operations, pivoting to selling software to staffing agencies, and ultimately being absorbed by two of the traditional staffing firms it set out to replace. BlueCrew raised roughly $150 million before being sold to EmployBridge, one of the largest traditional industrial staffing companies in the country. The marketplaces that did survive, Instawork, Wonolo, and Qwick, are real businesses, but look at what survival cost: nine-figure funding and large operations organizations doing the matching, vetting, and support by hand. Marketplace ambitions with staffing-agency cost structures. The graveyard of 2010s marketplaces is mostly companies that could no longer afford to be subsidized by venture capital, and the survivors are the ones that figured out how to keep the subsidies going.
What actually changed
AI agents made software capable of exactly the work that used to force marketplaces to hire: screening, matching, monitoring, pricing, and communicating. The unglamorous middle-office of a marketplace which ate my generation of founders alive, and that Uber and DoorDash spent billions brute-forcing with headcount, is now largely a models-and-workflows problem.
This is why the Shiftgig story matters. This generation of companies didn’t fail for lack of supply or demand. Restaurants wanted shifts filled and workers wanted shifts; supply and demand was never the problem. These companies failed because matching, vetting, and support took a coordinator’s salary for every increment of volume, and the venture funding was required to a bridge that never ended. Operations costs were the binding constraint. This is precisely the variable AI has changed and it is why this generation of the model deserves a fresh look rather than a pattern-matched dismissal.
One company in our portfolio, Croux, makes this concrete. On paper, Croux could have been B2B staffing software for hospitality, selling the scheduling tool and letting restaurants keep struggling to find people. Instead, it’s a marketplace matching hospitality talent with open shifts, and the entire operation runs with a small team. AI agents handle recruiting, matching, fraud detection, and pricing. Fill rates sit around ninety percent. The company’s proprietary trust score, built from its own transaction data, does the vetting that a 2015 version of this business would have needed a team to do.
Put differently: what used to require a DoorDash level fundraise, now requires a seed round. And that points to three structural reasons the SaaS-to-marketplace shift makes sense now.
- The economics flipped from seats to transactions. Picture one restaurant as a customer under both models. As a SaaS customer, it pays for a scheduling tool, call it $300 a month, $3,600 a year. No matter how busy the restaurant gets, that’s the upper limit of spend. As a marketplace customer, that same restaurant spends something like $400,000 a year on hourly labor. Fill even a quarter of those shifts at a 20% take rate and it’s worth roughly $20,000 a year — five to six times the SaaS ceiling, from a single location, growing every time the restaurant gets busier. Yes, a marketplace bears costs a tool doesn’t like payments, insurance, and support, so the fair comparison is gross profit, not revenue; but when agents are doing the coordination, the gross profit dollars from participating in the payroll still swamp the software fee. The tool sells into the software budget; the marketplace participates in the payroll. Founders always knew this math and many elected to stay on the SaaS side anyway, because the operations ate the margin, and the revenue was less predictable. That was the right call, until AI made the ops nearly free. Now, staying on the tool side of a transaction you could be operating isn’t prudence. It’s leaving the business on the table.
- The marketplace learns things the tool never finds out. The scheduling tool knows one thing about Saturday night: a shift existed and someone’s name was on it. It has no idea whether that person showed up, kept up during the rush, or got asked back. Monday morning, the tool is exactly as smart as it was last week. The marketplace knows everything, because it was in the transaction: who showed up, who was requested by name, what pay rate filled the shift in ten minutes versus the rate that left the shift sit unfilled for two days. Every shift teaches it something, and that knowledge feeds straight back into better matches and smarter pricing, which attracts more restaurants and workers, which produces more learning. It’s why a two-year-old marketplace can be dramatically better at its job than a ten-year-old tool. Everyone in AI claims a “data moat”; knowing how millions of real transactions actually turned out is one of the few that’s real, because a competitor can copy your features but can’t copy your history.
- Good AI finds the workers the old system couldn’t see. Hiring normally screens people with a résumé, a credential, a background-check box: crude filters that reject a lot of genuinely great people. The bartender with a two-year caregiving gap or the line cook with no certificate who’s never missed a shift can get overlooked. A marketplace with real outcome data doesn’t have to guess from paper: it has watched this person show up on time forty times and get requested by name at three venues. That track record qualifies people the old filters bounced, and every one of them is someone who now gets work they couldn’t access before. The tool made the existing market slightly more organized. The marketplace makes the market itself bigger.
That’s the mechanism behind the claim I opened with. The jobs AI absorbed at these companies, screening queues, scheduling tetris, phone-tag, mostly never existed to be lost; the realistic alternative was the market going unserved, because the operations were too expensive for anyone to build the network at all. The thing AI eliminated wasn’t employment. It was the toll booth between people and paid work. Displacement elsewhere in the economy is real and deserves serious treatment. But in this category, the honest description is the opposite of the doom framing: AI is doing the unpaid-for friction, and humans are doing more of the paid work.
The quiet generalization
Once you see the pattern, you see it everywhere. Care navigation that would have been a directory becomes a managed marketplace coordinating payers, clinicians, and contractors. A doula directory becomes a matching platform that owns the outcome. Benefits software becomes a navigation marketplace connecting people to dollars they didn't know existed. Creator tools become platforms that give creators actual options for how their money moves. In each case, the founder faced the same fork, build the tool or operate the network, and AI changed which fork is rational. The result, at least in our portfolio, is companies growing at huge multiples on seed-stage burn, with take rates and gross margins expanding while they're still small because the marginal cost of the next match is a model call, not a coordinator's afternoon.
The honest caveats
Not every SaaS business should become a marketplace. Stepping into the transaction means owning quality risk, liability, and the temptation of both sides to go around you. The categories where this works share a shape: genuine fragmentation on both sides, recurring transactions, outcomes measurable enough to feed a trust model, and a matching problem hard enough that the marketplace earns its take rate every time. If your customers would happily disinter-mediate you after the first match, you have a lead generation business with extra steps. Agent leverage is not the same as no operations — the best AI-native marketplaces still have humans, deployed at the edge cases.
Why I'm writing this
I lived the version of this industry where the advice was "don't fund marketplaces until they're big enough to survive their own operations." That advice is now wrong, and it's wrong in a way that creates opportunity for exactly the founders who used to be told to build the safer SaaS version of their idea.
If you're building software that sits next to a transaction, staffing, care, services, money movement, ask the uncomfortable question: is the tool the business, or is the tool the excuse? Five years ago, the honest answer was usually "the tool, because operations will kill you." Today, the operations are software too.
Sometimes I think about what my furniture marketplace would have looked like with agents running the warehouse scheduling, the logistics routing, the two-sided balancing act that consumed my team. We were right about the market. We were just a decade early to the fix.
The founders building now don't have to be. Because the operations were never the point. The point is the bartender with the caregiving gap on her résumé, working Saturday night because a model looked at forty on-time arrivals instead of a two-year hole. The point is the shift that got filled, the doula who got matched, the contractor who got paid to build a ramp outside a home. Marketplaces were broken for twenty years, and the whole time, that's what was waiting on the other side of the fix.
The machines took over the paperwork. The people got the work.
Shruti Shah is a General Partner at Symphonic Capital, a seed fund backing overlooked founders building AI enabled and AI native businesses in healthcare, fintech, climate. Before joining Symphonic, she co-founded a used-furniture marketplace — which is to say, she learned the operational lessons in this essay the expensive way. Symphonic's portfolio includes several of the AI-native marketplaces described above.