Elias Torres, Co-Founder and CEO @ Agency: What No One Tells You About Selling Your Company

Summary

• Torres calls selling Drift to Vista for $1.2BN in September 2021 “the biggest failure in my life” — not the price, but the failure to build something sustainable. 800 people (200 in customer success) killed product speed, and the soft culture is why there’s no public company: “I would have a public company right now if it wasn’t for all those carrots that I gave, and candy bars, and trophies.” The unspoken part of exits: “it’s sad to be nobody… you walk around in your own company and you don’t matter anymore.”
• The incumbent scorecard is brutal: “the elephants are not dancing, they’re trying to dance.” Google “just dropped the ball” — invented the Transformers paper, yet “everything they put out doesn’t work; NotebookLM, the great one, it’s already dead.” Microsoft’s OpenAI bet was “genius” but on Copilot “nobody’s using the product.” ChatGPT is “the king” — yet still operator-software, “not the real solution, not the future.” He concedes distribution and data are the incumbents’ biggest edges but insists “speed is everything.”
• The potential is 100-person companies doing billions in revenue — “smaller is better” was always true, “but because we couldn’t achieve billions in revenue without more people, we had no choice.” Agents could change that: “we’re just really trading people for compute.” Venture consequence he endorses: early rounds stay equity, growth financing becomes debt against predictable revenue, because far less equity gets burned on headcount.
• Adoption will be slower than the hype because “humans are the biggest blocker” — CEOs tell him yes, then delegate the decision down layers that say “next quarter.” Hence his contrarian call: AI services companies will be bigger than AI technology providers within five years (Paul Graham trashed it; Torres personally made $2M in 6-7 months consulting for OpenAI’s customers). Timeline: 5-10 years, not 12-24 months — he offered to bet Harry against sudden 10% GDP growth: “we just discovered fire.”
• Positions: the one stock he’d hold ten years is Nvidia (“I’m scared of the giants, they’re not going to last… we’re thirsty for compute”); his biggest public short is Salesforce; and he’s not selling Klaviyo, where he was first investor — “super underappreciated.” Counter-caveat on the new winners: Cursor-style revenue “could disappear — you can easily switch to Windsurf… this is a brand-new phenomenon.”
• Hiring doctrine from the man Hiten Shah, Dharmesh Shah, and Pat Grady call the best recruiter they know: hire slow, fire fast, paid contract trials before offers, intensity as the filter (“open up your computer, I want to see your Git right now”). He hired five Northeastern grads in one day off their friends’ GitHubs and had them shipping code the same day. Titles are dying; comp moves to profit share, revenue share, and higher salaries.
• The life thesis underneath: grit is manufactured by adversity — homeless at high-school graduation in Nicaragua, first tech on a stolen IBM in his dad’s garage. Have kids early (his first at 27, wife met at 17): “you can do both. It’s not going to be pretty, it’s going to be ugly” — but “I can die happy today.”

Deep dive

1. “I pretend I’m a nihilist to cope with the fact that I actually care way too much”

• The opening paradox sets the register: Torres’s “I don’t care” is armor, not apathy. Entrepreneurship’s rejection volume — people quitting on him, customers churning — taught him “you get beaten up so much that you just realize it’s going to be okay.” Money removed roughly “one-third of the problems”: broken plumbing, cancelled trips — “just pay it, move on.”
• Stebbings argues that the standard tell-your-21-year-old-self advice (“it’ll all be okay”) can remove the fear that drives people to succeed. Torres’s version: “things are not as hard as they seem.” Even frontier AI: tell a stranger they can build an LLM themselves and they’ll say it’s too hard — “it’s really not that hard, the code fits on a page on the screen.”
• The calibration comes from Nicaragua: he was three during the 1979 revolution, living in a house that wasn’t theirs, homeless at high-school graduation when the owners reclaimed it — saved by a grandmother’s green card. First computer: a stolen IBM his father bought and stuck in the garage in LA, where he typed homework in WordPerfect and Lotus 1-2-3. “That’s why immigrants do well in startups.”
• For his own kids he “manufactures adversity” — and takes the counterargument seriously: his daughter called out his “quit your job now” LinkedIn post — “not everybody has the luxury that you have today.” His answer: he didn’t have it either when he started.

2. Elephants can’t dance — they can sing, maybe

• The conviction dates to IBM in 1998-99, building browser apps in the group that put ibm.com online: a 400,000-person company can’t move fast. Against Gerstner’s famous book: “they claim they’re dancing, but they don’t. They can sing, maybe.”
• The scorecard, name by name: Google “just dropped the ball — they invented this Transformers paper and everything they put out doesn’t work. NotebookLM, the great one? It’s already dead.” Microsoft earns real respect — its early OpenAI investment was “genius… changes the world” — but Copilot benefits from incumbency: “they’re reaping the benefit of being early, but nobody’s using the product. It just doesn’t work.”
• The hierarchy as he sees it: “the king is ChatGPT” — the one product with daily adoption — “but that’s not the real solution, that’s not the future.” Copilot and Cursor are “a distant second.”

3. Harry’s pushback: if distribution wins, does speed even matter?

• Stebbings takes the incumbent side name by name: Google’s endpoints reach billions. Torres concedes cleanly — “distribution is the biggest advantage that the incumbents have”, plus the data — but holds the line on execution: it’s been about five years since GPT. “Is that fast?”
• Then the sharper question: if technology commoditizes, why not wait and pour it into your distribution? Torres, categorically: “Speed is everything. Speed is what creates the American dream.” Speed creates the opportunity for others to win.

4. Humans are the blockage

• Why elephants can’t dance: headcount itself. Companies with “a thousand product managers” where “nobody gets fired for saying no” — people coasting toward promotions, writing five-things-I-did-this-week emails. “You’re toast. You can’t move.” And his most-hated management ritual: “if my whole schedule is filled with one-on-ones, shoot me” — the worst message in any company is “Elias, can we talk?”
• At Agency (15 people) there is one meeting for the whole company: “show me what you did yesterday, show me what you’re going to do today. The rest of the time I’m selling or I’m helping customers.” At 50 people it becomes groups of ten owning products — “how do you think I built HubSpot?”
• The enterprise-sales reality check: CEOs tell him “yes, let’s get Agency in here right now” — then pass him down the org, where every layer resists, and it comes back as “I’ll get back to you next quarter.” “CEOs are making a big mistake… they should be owning that decision.” If that happens at 2,000 people, “what do you think happens at 4,000, 10,000, 20,000, 40,000?” Companies that don’t change fast enough “are not going to last.”
• The demand signal is already there: public-company friends “spending $30 million on CS — and it’s not even the money, I can’t hire fast enough and train them… If you could train an AI to do that job and just have straight compute do it 24/7, which one do you prefer?”

5. Software still assumes an operator — that’s the thing that dies

• His core product insight: the last 20-30 years of software was built like cars — the whole environment assumes a human drives. The future is driverless: “I just want to tell it where I want to go and when.” ChatGPT fails his own test — “ChatGPT still assumes that I have to use it” — though he grants the step-function framing: like levels of autonomous driving, and Operator is moving up the curve. He includes himself in the indictment: “I’ve been guilty of this. I built a lot of this software.”
• On the comfortable consensus that AI just makes everyone more productive without cutting workforces: flat rejection — that’s nonsense, and guests he “loves and respects” who predict headcount growth get an incredulous “what?!” A post about hiring a thousand new sales reps: “I’m sorry — if we’re building the promise that AI is going to do the work of humans…”
• What the humans do instead: serve the rest of the world’s niches — “why are we all focused on the same market?” — or genuinely different lives: “I’ll go have a farm and plant some beans and just watch them grow.” Learn to paint, study the pharaohs. “I’m not going to be like, let me take another job at Salesforce.”

6. 100 people, billions in revenue — and what it does to venture

• The historical frame: every generation — industrial, internet, SaaS — assumed larger was better, “but because we couldn’t achieve billions in revenue without more people, we had no choice. Every big company today started small.” Agents could remove that constraint: much smaller companies with billions in revenue is “the potential” — hedged honestly with “maybe it doesn’t work out that way.”
• Pressed by Stebbings on the venture model, he goes further than most founders will: smaller teams need less funding, and since “we’re just really trading people for compute,” far less equity gets distributed round after round. His conclusion: early financing stays equity, but growth financing becomes debt financing on predictable revenue. “Absolutely, of course.”
• The flip side of hypergrowth: Cursor “is great because it could grow so fast, unbelievable — but that’s a different style of product. You can gain it easily and you can lose it easily… you can switch to Windsurf, you can go back to Copilot. That revenue could disappear. This is a brand-new phenomenon and we do not know how it plays out.” His counter-strategy at Agency: “the distribution of lifetime customers” — B2B trust and data relationships, at scale, with a small team. Stebbings’s worry stands alongside it: a generation of SaaS companies benchmarked against Cursor “will not get funding.”

7. “Drift is the biggest failure in my life”

• Stebbings calls the framing a cop-out — you sold for $1.2BN — but Torres holds it: “I wanted to build something great… enduring, that the customers could keep using and getting value.” Why it didn’t transpire: “too many people. My problem, my fault” — 800 employees, 200 in customer success, and a product that wasn’t growing. “I let people down, I let my customers down. I don’t want to do that again.”
• The most self-lacerating line of the episode: coaches told him to soften, add carrots — “that’s why I would have a public company right now, if it wasn’t for all those carrots that I gave, and candy bars, and trophies.” His conclusion on adaptive leadership: “Good luck… I’m done changing for people.”
• Why it wasn’t salvageable: companies are “too big, too inefficient.” The survival prescription for today’s incumbents: “if they do what Twitter did — reduce themselves to 20%, build an amazing product, and are willing to break all the pricing models and revenues — then they will survive.”
• The timing wound: “I was so close to the LLMs being there.” And the benchmark that stings — HubSpot, “my thing,” is a $30BN company where Brian and Dharmesh still get to be involved. “That’s what I wanted to build, but I made mistakes along the way that didn’t create a foundation solid enough.”

8. The sale: unemotional in, depression out

• The decision itself he defends as “the smartest thing I did — not be emotional about it.” September 2021, peers riding the wave: Outreach’s Manny and another similar company getting “crazy valuations,” his own investors multiplying his equity — “you’re worth a billion now, you could be worth two or four.” He sold for $1.2 billion. His personal take: enough that “it changed my life.”
• What nobody tells you: “it’s sad to be nobody. You bring a new CEO, new management — it hurts the ego. You created this and you’re no longer relevant.” Then 7-10 months of traveling and “trying to find myself” — he cites a graph about depression and sadness after the sale. “But it was great, I needed that break.” Money’s role, from the quickfire: doesn’t make you happy, “but it definitely makes things easier.”
• The road not taken: a two-hour, many-bottles dinner with likely David Fialkow on how a VC firm works ended it — “no f—ing way. I’m a builder. I’d rather build, man — it’s so much easier, so much more fun.” He hates watching founders take journeys he can see are wrong.

9. Recruiting: intensity is the test, and everything is a red flag

• The criteria: high agency, IQ, grit, execution speed, obsession. Separately, people who ran psychology tests called his engineering org “the most extroverted engineering team I’ve ever seen.” The interview is deliberately abrasive: “open up your computer, I want to see your Git right now. Show me what you committed… bang bang bang.” Glassdoor reviews saying “I hated interviewing with this guy” are the feature, not the bug: “that’s the test — if they can handle me and they want to work with me, that is what has made the teams work.”
• The process inverted with experience: at Drift it was a numbers game and “I would convince more people than I should have.” Now: “hire slow, fire fast” (he used to hire fast, fire slowish), with part-time contracting in the process — “show me your grit, how much work can you do while you’re doing something else.” The mis-hire post-mortem is pure self-blame: “I saw it, but I was desperate… I convinced myself it was going to be okay.” To SaaS founders under quota pressure hiring the good-enough candidate: “I’ve been there. Don’t do it. It is not worth it.” And on references: “if you call me, you hear the truth” — the industry just recycles known-bad hires.
• The Northeastern playbook, as told: he saw one great student, asked “who’s your friend? Let me see that GitHub” — and made five offers in one day. Onboarding: “come to my house, we have a chef, we cook them lunch, we ship all the laptops to the house, and we just ship code that day.” The bar is explicit: two to three times last week’s output, “and whoever doesn’t make that goes — they know that coming in.”
• On CVs: he agrees “100%” that serial two-year bouncers are a red flag — then flips it: “everything is a red flag. If you stayed there for a long time, it’s a red flag too. What’s your slope?” Even title acceleration means little — “great people use data, not great people use stories.” Titles themselves are ego artifacts he expects to disappear; the comp rethink for a 100-person, billion-revenue company: profit sharing, revenue sharing, more equity, higher salaries — burn out after two years, “take your equity, it’s going to be worth millions, maybe billions, go start something else.”

10. The tradeable calls: services beat models, adoption runs long, long NVDA / short CRM

• Between Drift and Agency he “sweated it out learning LLMs” and made $2 million in six-seven months consulting — OpenAI sent him leads from customers it wouldn’t hand-hold (Red Bull, Ticketmaster, likely the NBA and likely Klaviyo), starting at $25k/month and scaling to $100k and $500k contracts. That experience underpins his tweet: “AI services companies will be bigger than AI technology providers in the next five years” — Paul Graham rubbished it, but the logic is his adoption thesis: “humans are the biggest blocker to adoption, and they’re going to need a lot of handholding.”
• On timelines he’s explicitly “on the longer side” — 5 to 10 years of transition, not 12-24 months: “there’s a lot of hype.” Industrial revolutions grew GDP at 1-2%; he doesn’t buy a sudden 10-20% and offered to put money on it against a 10%-GDP framing (“let’s drop some money on this”). The summary metaphor: “we just discovered fire — the rest is going to take time.” Most industries adopt slower than tech, and tech itself “is being extremely slow to adapt this technology.”
• The quickfire book: one public stock for ten years — Nvidia: “I’m scared of the giants, they’re not going to last, but… we’re thirsty for compute.” Biggest public short: Salesforce — “oh sh—, Benioff’s going to kill me” — and Stebbings notes everyone he asks says the same. And the private position he won’t touch: Klaviyo, where he was first investor and convinced Andrew to take VC money — “I still think it’s super underappreciated. That’s why I’m not selling my stock.”
• The post-ZIRP self-revision: “profitability was never a discussion — Pat gave me a class on free cash flow at one point.” Now: “I’m thinking more about an enduring company than anything else… I’m more driven by legacy now than I am about money” — profitable at 100 people, built to last, holding David Cancel’s lesson that the answer to quick-or-great is “always both.”

Verification Notes

• The captions render the VC-firm dinner guest as “David falo”; “likely David Fialkow” is an uncertain reading.
• The consulting customer renders as “clayo”; “likely Klaviyo” is an uncertain reading.