I brought together accounts from thirty-three founders and operators to explore where their experiences meet and where they diverge: hiring, org design, the cost of growing, and building with AI.
For SaaS founders and operators making consequential people decisions.
The accounts collected here return to difficult people decisions: the leader hired too early, the team built for a future that never came, the culture that drifted while attention went elsewhere. Similar questions recur, even when the founders draw different lessons.
On what to do about it, they split. Ask them and you get contradictions. One says never lower the bar; another says the great person you turned away is the most expensive mistake you will never see. One added layers of management to survive doubling every year; another cancelled four executive searches and took the reports himself. One built rural offices in Tamil Nadu; another moved headquarters to San Francisco.
So this piece is built around the contradictions. Part one sets out thirteen tensions: two credible positions, the founders behind each, one way to hold both, and a question to carry. The first three are for anyone who ran something large before starting something new, on a product where AI does much of the work. Part two tells the stories. Part three sets out questions this collection leaves open. There is no checklist here. Argue with it.
Hire for strength rather than lack of weakness.Ben Horowitz, co-founder of Opsware · the one verbatim line here; everything else is paraphrased and linked
Each pairs two positions serious founders hold at once. The “reading” under each pair is one way to fit them together. It is my interpretation, and open to dispute.
There is no such thing as a great executive in general, only a great executive for one company at one moment. Someone superb at scale can be wrong for a company that has not reached it. Lemkin’s ex-Salesforce VP of Sales tripled burn at EchoSign. Doshi and El-Hage both did the job themselves before hiring for it.
Horowitz (Opsware) · Lemkin (EchoSign) · Doshi (Mixpanel)Two defining SaaS founders came straight out of big-company sales. Benioff spent 13 years at Oracle and became its youngest vice president at 27. Halligan spent a decade at PTC and rose to senior vice president of sales. Both knew exactly how enterprises buy software and what frustrated them about it.
Benioff (Salesforce) · Halligan (HubSpot)ReadingWhat transfers from running something large is judgment: about customers, capital, people and how big organisations buy. The machine stays behind: the brand that opens doors, the teams that turn a decision into execution, the numbers that arrive without asking. The main risk is reaching for a machine that does not exist yet, and hiring the org chart of the company you left. Benioff and Halligan both did the early work themselves and paired with builders, which is the next tension.
Which of your instincts come from judgment, and which from having had a machine behind you?
The non-technical founders who built great software companies paired early with technical co-founders who held real authority. Benioff started Salesforce with Parker Harris, who designed its multi-tenant architecture. Halligan started HubSpot with Dharmesh Shah, whom he met at MIT.
Benioff (Salesforce) · Halligan (HubSpot)A founder can’t hand over judgment entirely. If you have never done a job, Horowitz says, bring in experts and write down what great looks like before you hire. Spolsky judges engineers by what they have shipped. Doshi tries every role himself first.
Horowitz (Opsware) · Spolsky (Fog Creek) · Doshi (Mixpanel)ReadingPairing and learning work together. The pairs that lasted split authority clearly, and the non-technical founder still learned enough to ask sharp questions. In an AI product the line blurs further, because how the model behaves is the product. My inference: a founder who can’t judge the model’s behaviour has handed away the product itself.
On the three decisions that most shape the product, who has the final say, and could you tell a good answer from a confident one?
In April 2025 Lütke made reflexive AI use a baseline expectation at Shopify. Teams now have to show why AI can’t do the work before asking for headcount, and AI use counts in reviews. AI-native companies report revenue per employee in the millions: Dealroom estimated about $3.3M at Cursor and $2M at Midjourney.
Lütke (Shopify) · Dealroom estimatesIn 2024 Klarna said its AI assistant did the work of 700 support agents. By May 2025 its CEO conceded that cost had weighed too heavily and quality had suffered, and Klarna started hiring people again. By 2026 he was describing human service as a premium.
Siemiatkowski (Klarna)ReadingBoth are asking where human judgment earns its cost. Lütke asks the question before a hire; Klarna answered it after the cuts and paid in quality. For a company that is AI-first from day one, the support functions may be where this shows up first. My inference: a small team with strong tools can run finance operations, recruiting coordination or support for much longer than the old ratios assume.
For each role you plan to add, what would the job look like if AI already did half of it, and which half must stay human?
Know precisely what the role needs before you meet anyone, ideally by doing it yourself for a while. Hire people who have done exactly this job at exactly your scale. Lemkin’s polished ex-Salesforce VP of Sales is the cautionary tale.
Horowitz (Opsware) · El-Hage (Drop) · Lemkin (EchoSign)People who can take anything from idea to shipped are rare, so find them and shape the role around them. Build teams around what each person does exceptionally well. Freshworks hired someone with no impressive CV, only a helpdesk they had built on GitHub.
Subramanian (Chargebee) · Rabois (Square) · Mathrubootham (Freshworks)Reading“Experience” hides two different things: the title and brand, and the stage. Lemkin’s VP had the title and the wrong stage. The Freshworks engineer had no title and the exact work. The camps may agree more than they admit. Both trust evidence of the real work. Both distrust credentials standing in for it.
For the seat that worries you most, is the playbook already known or still being written? And what evidence would actually change your mind about a candidate?
Only hire someone you would be glad to have ten more of. For early employees, ask whether they could have been a co-founder. If in doubt, don’t hire. Restraint under pressure is a founder skill.
Collin (Front) · Spolsky (Fog Creek) · Horowitz (Opsware)A bad hire is visible and can be fixed. A great person you turned down is invisible, and the cost has no ceiling. Ward’s example is Facebook passing on Brian Acton, who went on to co-found WhatsApp.
Ward (Carta)ReadingHow afraid you should be of a false positive depends on how expensive a mistake is to undo. Carta pays the same departure package whether the company or the employee ends the job, which makes being wrong cheaper and honest conversations easier. The companies that can afford a bolder bar usually made mistakes survivable first.
How costly is a wrong hire here really, counting the months it takes to admit it? What would make it cheaper?
Buffer went from 34 to 94 people in a year, including a customer success team for an upmarket move the product never made. Atrium hired to win customers before the product was right. Mixpanel chased enterprise to 350 people. All three regretted it.
Gascoigne (Buffer) · Kan (Atrium) · Doshi (Mixpanel)One of Freshworks’ deliberate bets was to invest in hiring when good talent was scarce. Mixpanel’s own co-founder later said they should have started hiring earlier, because finding the right people takes far longer than expected.
Mathrubootham (Freshworks) · Trefren (Mixpanel)ReadingThe regrets are about hiring ahead of the product. The bets that paid off were about hiring ahead of scarcity: building relationships and pipeline for people you will need, long before the seat opens. Perhaps pipeline early, seats late.
Of the roles you expect to add next year, which are bets on the product and which are bets on the talent market?
As Chargebee doubled every year, its leaders were swamped by daily decisions. Adding a management layer let them think in quarters and years.
Subramanian (Chargebee)With Buffer stalling, Gascoigne was running four executive searches at once. He cancelled all of them, took the direct reports himself, and told product and marketing they were already complete teams.
Gascoigne (Buffer, via a 2026 case study)ReadingThe direction of travel may decide it. In growth, a layer buys leaders a longer horizon. In a stall, the same layer puts distance between leaders and the work at the moment closeness matters most. Both founders were really deciding where the leader’s attention should go.
Sketch the reporting lines you would want if growth doubled next year, then the ones if it halved. How different are the two pictures?
Paras Chopra sat in on every Wingify interview until the company had about 120 people. Horowitz argues the CEO should make the final call on executives alone, because consensus drifts toward the mediocre.
Chopra (Wingify) · Horowitz (Opsware)The CEO’s job is to keep firing themselves from last quarter’s work. In a fast-growing company, anyone who wants to grow has to hand away what they built every few months.
Subramanian (Chargebee) · Graham (Facebook, operator)ReadingFounders tend to hold on to hiring judgment longest and let go of execution first. That seems right. Taste in people is the part of a founder that compounds; in execution, the founder becomes the bottleneck. The hard part is being honest about which decisions belong where.
Which of your decisions carry your judgment into the company for years, and which only feel as though they do?
HubSpot treats its Culture Code as a product and revises it about every six months. GitLab runs on a public handbook that answers most questions before anyone has to ask. Zapier wrote down the traits of its best people and built them into rubrics.
Halligan (HubSpot) · Sijbrandij (GitLab) · Foster (Zapier)GitLab’s own founder admits people still fall back on Slack and meetings instead of the handbook. Intercom was famous for its writing, and McCabe still returned to drift and infighting. Moz was known for transparency, and it didn’t prevent serious mistakes.
Sijbrandij (GitLab) · McCabe (Intercom) · Fishkin (Moz)ReadingThe written culture holds when it shows up in decisions people can see: who gets hired, promoted and let go. Zapier’s move was to put the same traits into performance reviews as into interviews. People read culture from those choices long before they read the document.
What would a thoughtful new hire conclude about the culture from the last five people promoted?
Zoho grew without venture money. Gumroad cut 75% of its staff and became profitable at a fraction of its size. Chargebee’s founders saved part of their salaries before starting and ran frugally for years.
Vembu (Zoho) · Lavingia (Gumroad) · Subramanian (Chargebee)Freshworks, Postman and Chargebee itself later raised large rounds and scaled globally. Capital bought speed and reach that bootstrapping could not.
Mathrubootham (Freshworks) · Asthana (Postman)ReadingFunding may matter less than whether spend is tied to evidence. Doshi blames overcapitalisation for Mixpanel’s sprawl. Buffer’s post shows how support cost builds up: salaries over 80% of spend, weak finance and HR, and no agreed floor for cash. Sacks’s burn multiple (cash burned per dollar of new recurring revenue) exists to make that connection visible.
If you had to justify every role outside product and sales to a sceptical outsider in one sentence each, which sentences would be hardest to write?
Zoho takes jobs to where people live: 25+ rural offices and a school that trains and hires students after Class XII. Freshworks acted global from Chennai with ten people, working US hours and rewriting its copy so it didn’t read as Indian.
Vembu (Zoho) · Mathrubootham (Freshworks)Postman moved its headquarters from Bangalore to San Francisco in 2017. MoEngage found no single person could open the US and built a small founding team there, with about $1M set aside over two years for presence and community.
Asthana (Postman) · Dodda (MoEngage)ReadingThey disagree about where headquarters belongs. They agree that the customer’s market needs a real team on the ground, which no single heroic hire can supply. My inference, which none of them states: talent outside the big metros is underpriced, and Zoho’s model looks partly like a bet on that.
Where does the company’s centre of gravity need to be for customers, and where for talent? Are those the same place?
Kan’s advice after Atrium was to start remote: the pool is wider and the money goes further than in San Francisco. Zapier has never had an office and grew across 17 countries.
Kan (Atrium) · Foster (Zapier)Most companies copy office habits onto video calls and stop there. The ones that get it right judge output and rely on writing. Managers used to watching presence are the biggest obstacle. Shah adds that trouble on a remote team, such as avoided conversations, is harder to see.
Mullenweg (Automattic) · Shah (KISSmetrics)ReadingDistance hides problems. Teams that write well and measure outcomes do better apart; teams that avoid hard conversations do worse, and later. Across India and the US there is a further tax: the overlap window. Asthana describes a working day split across Bangalore, San Francisco and Austin.
Honestly, which of Mullenweg’s five levels describes how work happens across time zones today, and who would disagree with your answer?
Early people carried the company when nobody else would. Chargebee was built by classmates, a college roommate and a trusted mentor. Zapier’s first hires were former colleagues. Trust built that way is hard to replace.
Subramanian (Chargebee) · Foster (Zapier)People who shine from zero to one often don’t carry into one to ten. A hypergrowth company cannot develop executives fast enough, so it has to hire leaders who are already ready. Halligan’s rule of thumb: look for people who were at your stage within the last five years.
Subramanian (Chargebee) · Horowitz (Opsware) · Halligan (HubSpot)ReadingThe kindest organisations say this out loud early, so a change of role reads as a stage of the company. Graham’s image of giving away your Legos and Ward’s symmetric exits both try to make moving on feel normal.
Who on the team is building the company they will want to work in next, and who is building one they will outgrow, or be outgrown by?
The episodes behind the tensions, kept short. Each links to the founder’s own telling.
He spent 13 years at Oracle and became its youngest vice president at 27. After a sabbatical in 1996 he founded Salesforce in 1999 with Parker Harris, Dave Moellenhoff and Frank Dominguez; Harris built the multi-tenant architecture. He has said no venture firm would fund the idea at first.
Salesforce history · CNBC · Behind the CloudHis April 2025 memo made reflexive AI use a baseline expectation, asked teams to show why AI couldn’t meet a need before requesting headcount, and added AI use to performance reviews. He published it himself after learning it was about to leak.
Lütke’s post · CNBCIn 2024 Klarna said its AI assistant handled two-thirds of customer chats, the work of about 700 agents. In 2025 the CEO admitted cost had driven the decision too much and quality had dropped. Klarna began hiring human agents again, now through a flexible remote model, while AI kept the simple cases.
CX Dive · ForbesHis first sales leader came from Salesforce, sounded right, and hired reps fast. Revenue fell and burn tripled. The replacement was an early customer who had built sales at LinkedIn, and sales doubled in about two months. Lemkin reckons most founders make the same first hire.
SaaStrBuffer nearly tripled headcount in a year. Salaries passed 80% of spend, nothing tied new hires to revenue, and cash fell to $1.3M with about five months left. Ten people were laid off. Gascoigne’s diagnosis was missing a financial lens on decisions, along with weak finance and HR and no agreed floor for the bank balance.
BufferWith the company stalling again, he stopped four executive searches, took the teams himself, and told them they were already complete. He also concluded Buffer had been forcing a premium, high-touch model onto a business built for volume.
Business of Software, 2026A case study about Gascoigne, written by Business of Software.
He estimates first-time founders get more than half of their early executive hires wrong. After one of his own, he made himself interview at least thirty people before the next offer, and let only people whose judgment he trusted sit on the panel, each with a veto.
First Round ReviewAtrium raised $75M to rebuild the law firm with software and shut down in 2020. Kan’s lessons: they hired before the product was right, culture never had time to form, and running a services business meant managing many people for little reward. He would start with a clear mission and go remote from day one.
Kan’s thread · Global Legal PostHe set out to build a billion-dollar company, laid off about three-quarters of his staff in 2015, and ended up running a profitable business with a tiny team. His essay struggles with being called a failure while serving millions of customers.
Reflecting on My FailureMixpanel moved upmarket into long enterprise deals and grew past 350 people before a near-death stretch around 2017 to 2019. Doshi points to overcapitalisation. His personal rule: don’t hire for a role you haven’t tried yourself.
Seven temptationsKISSmetrics was early, well funded and loved by customers. Shah blames his own distraction: new ideas after product-market fit, management by dropping in, and letting people stay too long while colleagues talked about them behind their backs.
My Billion Dollar MistakeHe stepped away as CEO for health reasons and returned to stalled growth and infighting. Weeks later ChatGPT launched. He bet the company on AI, deliberately walking away from about $60M of revenue to do it.
First Round podcast · Lenny’s NewsletterMoz was famous for openness. Fishkin’s book still lists hires made before he understood the roles, the pressures of venture expectations, and how his own arrogance and depression shaped decisions.
Lost and Founder · SparkToroWard treats every hire as a sign the organisation didn’t create enough leverage, and argues the invisible cost of rejecting great people outweighs the visible cost of bad hires. Carta pays market-banded salaries, makes one best offer with no negotiation, and gives the same departure package whoever ends the job.
How to Hire · MediumPatrick Collison’s advice to her: for each early hire, ask whether they could have been a co-founder. She came to see Front’s success as a matter of discipline, including the discipline not to hire when a role felt urgent.
First Round ReviewWhen Zendesk raised prices in 2010, he offered its unhappy customers the rest of their contracts free. From Chennai he had the team working US hours and acting global while still tiny. One key early hire had no impressive CV, only a helpdesk they had built on GitHub.
Lessons from All In · SaaStrZoho calls it transnational localism: rooted locally, connected globally. It runs a hub-and-spoke network of more than 25 rural offices, and Zoho Schools trains students straight out of Class XII and hires them as employees.
Zoho Rural Revival · diginomicaThe founders were classmates, a roommate and a mentor from Zoho. When the money ran out, he asked his wife to sell a rental flat rather than quit. Later, doubling every year, he added a management layer so leaders could look beyond the week, and describes the CEO’s job as continually firing yourself.
Insight Partners · The KenHis view on entering the US from India: experienced sellers run existing playbooks, and the first US hire has to write one. He built a small founding team there and set aside roughly $1M over two years for community and co-marketing.
SaaSBoomiHe has watched Indian founders hire their first US seller on quota records from late-stage companies. One such hire struggled without an established process to lean on; a less experienced but more resilient seller did better.
ECL guidePostman began as a side project at Yahoo Bangalore. Some of its earliest community supporters later joined as engineers, and the first marketing team grew from community work. Headquarters moved to San Francisco in 2017, and Bangalore stayed a core base.
Postman blogHe sat in on every interview until the company had about 120 people, looking for candidates as obsessed with the product as he was. The company was bootstrapped and profitable.
VelocityBefore HubSpot he spent a decade at PTC, rising to senior vice president of sales, then co-founded the company with Dharmesh Shah, a technical founder he met at MIT. A CEO peer group later convinced him that culture is how a company scales: how people decide when the leader isn’t there. HubSpot’s public Culture Code is revised roughly every six months. He also warns against hiring from companies far bigger than yours.
Lessons from HalliganThe only time Zapier’s team shared a building was during Y Combinator in 2012. Early hires were former colleagues. Later, Zapier named the four or five traits of people who do well there and used them in interviews and performance reviews alike.
Interview · Cadence talkHis ladder runs from office habits moved online, through async-first working, to judging people purely on output. Individual contributors climb easily, since many have freelanced. Managers used to watching presence hold companies back. Distributed companies, he notes, end up hiring for writing.
ma.ttGitLab runs on a public handbook meant to answer most questions without anyone asking. Its founder’s candid admission: the hardest part is getting people to update the handbook instead of falling back on Slack, email and meetings.
GitLab handbookShe watched Facebook grow from 500 to 5,500 people. Her image is a child asked to share the Lego tower they started: people in a fast-growing company have to hand off what they built every few months, and it hurts more than anyone admits.
First Round ReviewGraham was an operator at Google and Facebook. She is included because nobody describes this better.
These accounts leave me with further questions. The gaps below describe the material reviewed for this collection.
Founders write “we hired too many people.” The accounts reviewed here do not give a detailed breakdown of finance, HR, legal and IT headcount against revenue over time. The best figures come from benchmarks (general and administrative cost at roughly 15% to 20% of revenue in SaaS, per OPEXEngine). Perhaps overhead gets noticed late because no single hire ever looks wrong.
Freshworks, Chargebee and Zoho all ran from Chennai, yet the accounts collected here do not explain which support functions sat in India, which sat near customers, and what that choice cost or saved. It is one of the most consequential org-design decisions a cross-border company makes, and one I would like to explore in more depth.
Long notice periods, counter-offers, uneven understanding of stock options and fast attrition in the big metros receive little attention in these accounts. Advice built in Silicon Valley assumes a hiring market that works differently.
If AI absorbs much of the entry-level work, the traditional apprenticeship of support, coordination and first drafts gets thinner. Every tension above assumes a supply of experienced people. Nobody in this set has written about how an AI-first company grows its own senior talent when the junior rungs of the ladder have partly gone.