Flexport CEO: Two Questions Every Founder Needs to Ask

Summary

• Flexport breaking news: $450M net revenue run rate, basically breakeven this year, up from $350M last year (~30%), with $600M penciled next and a goal of “30% every year for 10 years.” Peterson applies Paul Graham’s two questions — is growth a hack, is the market big enough — and answers “no hacks, it’s actually a grind” plus under 1% share of a market that’s 11% of GDP. IPO comes at “a few hundred million of likely EBITDA… it could happen in a couple years.”
• Enterprise AI spend is real but self-limiting: Flexport pays ~$5M/year for LLMs (doubled in a few months, no budget cap on the Anthropic contract), is agentifying ~100 core workflows (5 live, 95 in development) — but “once they’re automated, I should stop spending the money on Anthropic,” and mundane work migrates to “basically free” open source. Stebbings’ math on the bull case: a likely Benioff figure of $300M Anthropic spend is 3.8% of developer salaries; trillion-dollar lab valuations need 18–20%.
• The tail risk Peterson actually fears is being cut off, not concentration: a lab deciding its compute “is more valuable for training superintelligence than it is for letting customers use it” — “we all just go back to being idiots we were two years ago.” Forced to pick one at equal price, he takes Anthropic over OpenAI: the enterprise business plus “a cohesive team that’s been together forever.” He recently invested at around 600 or something — “it already ran away.”
• The SaaS shakedown is starting: “I think selling SaaS to tech companies is going to be a tough business cuz we can build stuff for ourselves.” His procurement team is building a PowerPoint case study of each SaaS Flexport replaced, then going down the vendor list — cut rates or “I’m just going to have to vibe… replace you guys.” Expected concession: “you get like 20% out of almost everybody.”
• VC herding is structural, not stylistic: the job is so good that the game becomes not getting fired, so partners channel-check everything — “most VCs actually collude more with competitors than with their own partners.” Corollaries for founders: never test the market with one or two meetings (cross-firm associate roundups tank you), never share your metrics, and per PG, “hear the no but ignore the why.”
• Remote work is “white collar fraud”: done honestly, WFH is labor arbitrage — an off-the-charts-IQ assistant in the Philippines at ~$500/month, “not the guy who’s making 250k a year and lives in Jackson Hole and wants to go skiing for 4 hours a day.” Flexport is 5-day in-office and Peterson is now re-concentrating leadership in SF; moving his CFO there “made the business much better.”
• “Revenge and patriotism is a great investment thesis” — wronged second-time founders (he was Rippling’s first investor; he also mentions an unclear Dario/Parker reference) are his best pattern. Angel math after ~200 checks: mark every check to zero at signing; a 3x has “zero relative impact” beside a 500–1000x — so founders shouldn’t grind a decade for 1.5x.
• Masa led Flexport’s $1B round after roughly an hour at his Woodside house, having Foxconn called live mid-meeting for diligence. His advice — be 10% cheaper than everyone, and if matched, go 10% cheaper again — was “a terrible strategy. I did not do that. We would have burned so much money.” Yet Peterson’s biggest change of mind in 12 months points Masa’s way: become the low-cost leader — “I think I was lying to myself.”

Deep dive

1. The VC job is too good — herding is the rational response

• Peterson’s inversion: list the job’s attributes — “pays really well,” no boss past partner level, no fixed schedule, and “it’s very hard to measure if you’re good or not on any reasonable time frame.” A non-owning partner can get fired, so the whole game becomes avoiding it: no scandals, and never letting your own partners think you do dumb deals. Hence consensus, channel-checking, and “most VCs actually collude more with competitors than with their own partners.”
• Peterson’s example of likely Keith Rabois: Rabois sense-checks deals with friends, and if they don’t think a deal is stupid or crazy, “he’s not doing his job.” Stebbings extends this by suggesting Founders Fund and contrarian people may escape the trap — some firms, some contrarian bones.
• Stebbings’ extension — richer investors make better investors: Sequoia is “focused on upside maximization,” not fearful of LPs or deployment pacing. Peterson’s twist is the inverse: his risk permission came from being cheap — at 25 he lived in second-tier China on $120/month rent, total opex ~$250 — “I knew I could always make 500 bucks a month… it gave me permission to take some risk.”
• What drives him: the fear of losing — “I don’t want to be a loser.” By his own founding standard he’s won (his first and only financial model reached $1M of revenue; “we did over two billion last year”), yet it doesn’t feel like winning. His “number” was $20M — 5% interest, $1M a year for doing nothing — but “I haven’t been that motivated by money. I’m more into power. I want to do big things.”

2. Flexport by the numbers — and the pivot to low-cost leader

• Breaking news on the show: run rate of ~$450M net revenue this year, basically break even, versus $350M last year — call it 30% — with “I think we’ll do 600” next and a goal of 30% a year for a decade.
• His filter is Paul Graham’s two questions for any hockey stick: is there an unsustainable hack, and is the market big enough to keep going? Flexport: “there’s no hacks, it’s actually a grind” — salespeople in the field, calling on businesses — and the market is enormous, share still under 1%.
• On exits: the word implies stopping, so he ignores it; Flexport intends to IPO “nicely profitable, making a few hundred million of likely EBITDA… it could happen in a couple years.” The “IPO window” concept baffles him — “what if you went public and then your price went down? Is that better?” — and if undervalued while generating cash, buy back your own shares. Envying Anthropic-speed revenue is a deadly sin: “you certainly should avoid it.”
• His biggest change of mind in 12 months: a decade of refusing to compete on price, now “pretty convinced that we need to be the low-cost leader” — “I think I was lying to myself because it was too hard to automate the work.” 2026 is a success only if Flexport hits its numbers and ~80% of the 95 in-development AI workflows (5 of ~100 are live and saving money) actually land — “otherwise we’re just spending money and not getting much back for the AI.”

3. Agents finish what RPA couldn’t — freight is email forwarding

• The business case: freight forwarding “should be called freight email forwarding” — people passing PDFs and moving data between ERPs, every customer with bespoke rules (notify me 10 days before my container lands; no, seven). Flexport’s tech is a giant if-then rules engine that “can’t really keep up,” leaving humans to manage the logic. RPA went a long way; “agents can go all the way” — full end-to-end automation. And most of the economy looks like this, not like Uber.
• Spend today: ~$5M a year on LLMs, doubled in the last few months, with no budget constraint on the Anthropic contract. Mercor’s founder (per Stebbings) spends more on models and compute than on salaries; Peterson isn’t close, “but there’s a world where that becomes” the norm. $20M a year in five years? “Yeah, probably… on LLMs.”
• Stebbings’ valuation math — worth keeping: a likely Benioff figure of $300M/year Anthropic spend works out to 3.8% of developer salaries; trillion-dollar valuations need 18–20%. Peterson accepts the burden of proof: “even at five [million], I better save 5 million a year in labor cost — or make my product way better.”
• The risk that actually rattles him: a false “your organization has reached your limit” email last Thursday produced a genuine shock — “a realistic scenario” that labs decide compute “is more valuable for training superintelligence than it is for letting customers use it” and cut everyone off: “we all just go back to being idiots we were two years ago.” Maybe governments step in. Concentration per se doesn’t bother him — “as long as I’m allowed to use OpenAI and Anthropic… they deserve it.”

4. Diminishing returns to frontier — open source takes the mundane, China takes the blame

• The shared conclusion: frontier models stay for coding and product surface, but an automated workflow running on a “basically free” open source model needs no upgrade — “there’s diminishing returns to frontier labs,” and once the manual work is automated, “I should stop spending the money on Anthropic.” Stebbings draws the market implication both agree on: “the core business is much smaller than you think.”
• On CCP-funded open source powering early Silicon Valley (the likely Keith Rabois fear Stebbings relays): “I don’t lose sleep over it… if they’re open source, they’re open source. Who cares where they’re from — we can use them.”
• On China itself — he lived there years and speaks passable Chinese: mutual US–China dependence is “underplayed,” interests more aligned than credited, and the saber-rattlers are “casually throwing around war between China and the United States without realizing that such a thing would be a nuclear war and you’d all be dead.” He puts the odds “pretty low.”
• Forced to invest in one at equal price: Anthropic — “no offense, cuz Sam’s a friend of mine” — after agreeing that its enterprise business seems great and citing its cohesive team, versus OpenAI’s departures. He recently invested at around 600 or something (“it already ran away”), sized so that “Anthropic go to zero, I don’t care” — and if intelligence gets really cheap, “that’s much better for the world.”

5. The SaaS shakedown: 20% off or get vibe-replaced

• The headline call: “the negotiation that we’re going to have with Salesforce is going to be a lot different than the last one… selling SaaS to tech companies is going to be a tough business cuz we can build stuff for ourselves.” Flexport spends a few million a year on Salesforce, has already fully replaced one or two SaaS products, “and more every few weeks.”
• The playbook, verbatim: procurement builds “the PowerPoint case study of the SaaS that we did replace and how we did it and how long it took,” then goes down the vendor list — reduce rates “or else I’m just going to have to vibe… replace you guys.” Expected haircut: “I think you get like 20% out of almost everybody.”
• The proof point Stebbings supplies (likely Curative’s founder): sold $5B of COVID test kits, pivoted to a $1B+-revenue health insurer, and replaced $600k/year of Salesforce with an in-built tool in 3 weeks.
• His own limits: “some cases it might be a bluff” — he won’t burn core engineering replacing SaaS (“I want to replace expensive labor and build our product”), and security paranoia is constant: “almost all of our competitors have had major hacks” and ransoms. The wrinkle: Slack is sticky — “nobody wants to build their own Slack” — so cutting Salesforce may just mean “they raised the rates on Slack.”

6. Remote work is white collar fraud

• The line in full: “I say it’s white collar fraud. I have a three-year-old and a 5-year-old. The idea that I could do any work at my house is like a total fantasy… there’s no work getting done at that house when the children are around.” Flexport is 5-day in-office as baseline; he let COVID-era remote linger “way too long and our culture suffered.” Was reinstating it hard? “Difficult for the people who don’t work here anymore.”
• His advice to founders scared of a revolt: “you don’t want to ever be afraid of your employees… they want a leader who’s going to go in the direction that they believe in. Even if they disagree… and if they opt out, that’s fine.”
• The honest version of WFH is labor arbitrage: his brother’s Philippines-based assistant has “off the charts IQ” at ~$500/month. “That’s who’s going to benefit from a work-from-home environment. It’s not the guy who’s making 250k a year and lives in Jackson Hole and wants to go skiing for 4 hours a day.”
• The reversal underway: SF went from 100 to ~75 people — 4% of the company — and “we maybe overdid it”; he’s pushing leadership back, and moving his CFO to SF “made the business much better.” Stebbings says he barely does Zooms (“the quality of interaction is so low… I’m so disengaged”) and is interested in Google’s hologram “dimension” units — at $250k each across 40 offices, Stebbings rules: “you can’t afford it.”

7. Headcount flat, boots on the ground, and the forward-deployed moat

• At ~2,000 employees, in 4 years the team “might be about the same” — the mix shifts from manual operations into customer-facing sales and account management, and “if the people are not able to make that shift, then we’re going to have to rebalance.”
• Stebbings pushes the super-contributor thesis (ClickUp’s likely Zeb Evans: cut people, pay the AI-leveraged stars 10x). Peterson wants those people but is “somewhat skeptical”: sales is a relationships game with a per-human ceiling AI might only double, and “if Yelp can have 3,600 salespeople, I feel like we should have…” The market math: logistics is 11% of GDP, physical-goods companies spend ~5% of revenue on it — “logistics is a much bigger market than software.” Flexport’s largest customer spends ~$150M/year; a dozen spend over $50M.
• The moat he envies: a competitor has 130 badged full-time logisticians inside a hyperscaler running its logistics — “we cannot crack the door… it’s an insanely great competitive advantage. I’m jealous.” Then the revenge win: that competitor got acquired, laid off the forward-deployed logistician who ran the customer’s RFP (and always chose her employer); the customer hired her, and she picked Flexport “to get back at them” — one of its biggest wins ever. “Never underestimate corporate incompetence.”
• On HR in the AI era — Stebbings’ proposal, which Peterson calls a good idea: humans aren’t a minable resource (“the employees are the company and the company is the employees. Everything else is a fiction”), but agents are — so the new function is “agent resources.” Stebbings then says “996” is being replaced by “007: midnight to midday, 7 days a week.”

8. Two fundraising screwups and the Founders Fund bailout

• Series B, ~2015–16: a good-but-unfamous firm preempted at $50M on a $500M valuation. Peterson decided he could get better terms from a bigger name, spent a weekend pitching with no data room, word got back, and the offer evaporated. The best offer from his salvage process: $275M with board control. He confessed everything to Founders Fund — and Peter at Founders Fund offered $300M without board control: “I told him he could have it at 275 and he offered 300… he didn’t need to do that.”
• On always taking the highest price: his historical instinct was to take the highest price (“Brian, likely Chesky, says don’t ever — that I disagree with”), but he concedes Stebbings’ one piece of generic advice: tier ones are worth a discount — being a Founders Fund company pulled in later investors, and founders “consistently underestimate” how much operators join companies off VC signal.
• The Founders Fund origin story: a likely Sam Altman — one of Flexport’s earliest investors — made the intro. Against Zero to One’s six-or-seven questions (he couldn’t remember which), Flexport aced five; on “small market,” Peterson admitted his monopoly framing would be fake, and Peter stopped him: “don’t be too dogmatic. It’s okay to have a big market.” A few weeks later Peter emailed asking to invest. What amazes him since: “how far in the future he’s able to look and be right” — plus they talk “two times faster than when we talk to other people.”
• The pitch he walked out of: third meeting, expecting a term sheet, the firm brought its own report sizing Flexport’s market at $6B. His co-founder’s line: “oh, so it’s smaller than the market for USB cables.” Peterson left “to take a phone call” and never came back — and Stebbings’ rule stands: “if you need BCG to convince them, just don’t bother.”

9. The rumor mill: never test the market, never share metrics

• Peterson once told a mid-level investor he was “going in a different direction” just to let him off the hook — within an hour, three other funds called asking who led. “The amount of rumor mill and collusion that happens in VC, founders have no idea… you’re looking for an outlier who will value you more than everybody else,” and the grapevine kills outliers.
• Stebbings supplies the mechanics — worth keeping: cross-firm associate WhatsApp groups and weekly roundups, because “the associates are more in business with that class than they are with that firm.” One lukewarm meeting and you’re in the roundup as “only at a million, not growing that fast” — and everyone else skips you. Hence: never dip a toe in with one or two meetings.
• Corollary: founders should basically never share their metrics — cherry-pick whichever number looks best, because “once you’ve shared a metric, you’re now committed to that being the metric.” And raise when confident (advice he credits to likely Mathilde Collin, from Front): “confidence emanates.”
• On rejections, PG’s rule: “you should hear the no but ignore the why” — the odds of investors being honest times the odds of being right are too low for signal. Peterson’s no-BS exception was: “I just didn’t think you told a very compelling story”; his reality was harsher — “the vast majority” of investors simply ghosted him after seemingly great pitches.

10. Angel power law, the revenge thesis, and Masa’s $1B hour

• He angel-invested from inside his YC batch — 13 of ~50 companies — because peers told him “the real story.” It wasn’t obvious who was great, “it was obvious who was terrible… if you could eliminate the bottom half you’re doing pretty well.” Hits across ~200 checks: first investor in Rippling (big check), Algolia (sold, made money), early Bitcoin (sold — “don’t kidnap me”). The miss: Cruise, in his own batch — Kyle “throwing this camera setup on the roof of his Honda Civic… I was like, this is not legit.” Stebbings’ counter-heuristic: “the crazier the thing, the more I’m like I need to do that.”
• The power-law lesson he wants founders to absorb: delete a 3x from his spreadsheet and it has “zero relative impact” on returns next to a couple of 500–1000x’s — “when I write a check, I just mark it to zero.” So don’t spend 5–10 years grinding out a 1.5x; Stebbings, flatly: “return 0.6x, it’s better. I don’t even want the investor updates.”
• The thesis, verbatim: “revenge and patriotism is a great investment thesis” — second-time founders “who feel like they were wronged the first time around,” citing Rippling and an unclear Dario/Parker reference.
• The Masa story: pitched at his Woodside house (“you can look up the Zillow estimate”), the meeting ran about an hour — “considering the size of the check, surprisingly short” — for the $1B round SoftBank led, with Masa’s aide calling Foxconn live mid-meeting for diligence. A Napoleon painting hung behind him; Peterson hunted for a Duke of Wellington painting as a troll gift and never found one worthy. Masa’s push: “whatever the price of freight is, you just be 10% cheaper than everybody. And then if someone matches you, you just be 10% cheaper than that — which is like a terrible strategy. I did not do that. We would have burned so much money.”