Kareem Amin

The through-line

Everything Kareem Amin says about Clay traces back to a single ambition that predates the product by years: "how do we give the power of programming to more people" [1][2]. He and his co-founder Nikolai came at it from the conviction that in the 21st century "the material is computation," the way oil or steel once was, and that anyone who cannot tell a machine what to do is locked out of the value being created [1]. They considered reinventing the terminal, decided that would only speed up engineers, and went looking instead for a group of non-engineers with more ideas than they could execute. That search landed on sales and marketing, and the name Clay came out of it: go-to-market work, but now you are an engineer [1].

The second conviction follows from the first and is the one he defends most often against the rest of his market: go-to-market is a creative act, not a funnel to be processed. Because it is creative, the tool has to stay open-ended rather than hand you an answer [1][4]. When LLMs arrived in 2022, he is careful to say they did not cause the takeoff. Clay was already growing, and the reason models slotted in so fast was that the company had already been built as an integration-first, coding-like product with go-to-market primitives, so the models "just boosted everything" [1]. His framing is that you set yourself up pointing in the right direction so that when the wave comes you go with it and are not fighting it [1].

On go-to-market as a creative act

The competitive claim is not a feature. Asked where the magic is, he says the advantage "is not in one single thing that we do" but in the approach: everyone else treats go-to-market as a funnel to process and automate, which he considers necessary but not sufficient [4]. He prefers to think in moments, borrowing activation energy from physics: water bubbles for a long time before it phase-shifts at 100 degrees, and a prospect accumulates moments of delight before becoming a customer [4]. He has a set of what he calls laws of go-to-market: you have to be unique, because doing what everyone else does is the baseline, and success itself destroys tactics, since scaling something raises the baseline and it stops working, so you must always be changing what you do [3]. The point of the product is to find "go-to-market alpha": "how are you different than everybody else? Otherwise, it's just noise" [1].

The analogies are consistent. "If figma is for designers Clay is for go to market teams" [2], and Photoshop or Figma to growth marketing and RevOps teams [5]. His sharpest version contrasts a microwave, where you set a time and press start and it works, with a guitar, six strings that look simple and take a lifetime, where playing teaches you music as you go [4]. He is explicit that this was counterintuitive: competitors position against Clay by saying salespeople are "coin-operated" and need something that just gives the answer, and Clay built the powerful tool instead [1]. The mission he repeats is that any idea for growing your company should be executable in Clay quickly [3][4][5], whether that is a firm selling to businesses with a lot of garbage using satellite view to spot accumulation [1], a devops company monitoring Clay's own status page and pitching them when it went down [2][4], physical mail, gifts, personalized websites or ads [2], or Verrata generating personalized landing pages for inbound visitors [4].

On the go-to-market engineer

He rejects both the copilot-for-SDRs path and the fully automated path in favour of what he calls a third way: take the job to be done, centralize it with someone more technical such as a RevOps person or growth marketer, and let them automate the automatable pieces [4][5]. The GTM engineer is a role he says Clay created as an AI-native evolution of RevOps, defined by treating go-to-market like an engineer treats systems: do we have the right data, did the tactics work, improve or do more of them [3]. It is not a demand for reorganisation. The role already exists, and advanced AEs and SDRs pick it up, so the approach is to meet people where they are and let results do the rest [2]. The incentive is blunt: people are making money and getting promoted using Clay, and that is why they like it [2].

He sees the same logic reshaping org charts, with sales, marketing and customer success converging into one go-to-market organisation mapped to prospect, lead, user and customer, supported by a GTM Ops team that runs the tooling [5]. Clay wants to be the system of action in that structure, sitting alongside the CRM and warehouse as systems of record [4]. Around the product an ecosystem has formed that he treats as evidence: a Slack community he puts at over 177,000 people, bootcamps teaching the tool, and Clay agencies of typically two to four people, some past a million in run rate within six months of starting [4]. He has also discussed the path to product-market fit in terms of building vertical, creating power users and founder psychology [8], and how the company scaled to $30M ARR in two years through product-led growth [6][7]. Early decisions he calls non-obvious included starting with outbound, targeting agencies rather than startups, and building the community in public [1].

On why the fully automated SDR falls short

He is precise about where LLMs already earn their keep: account research, contextual search used for scoring similarity, summarising calls and emails so nobody types data into the CRM by hand, extracting facts from websites, and generating short snippets and value props for outreach [5]. Claygent is an implementation of the ReAct paper, an agent that plans and executes across scraping, summarisation and Clay's own company and people tools [4]; Navigator extends that to browsing the web like a person, filling forms and navigating older or government sites to get information that is not neatly compiled [3]. Keeping the use case narrow, research about companies and people, is what lets them build meaningful evals and know whether they are improving [3].

What he does not believe is full autonomy. A model can generate a plan for "find me more customers like these," but the space of decisions is large, plans get commoditized quickly, feedback is hard to route back in, and any breakdown in a multi-step chain "is a disaster" [5]. Underneath sits his argument about attention: human brains are very good at sorting crap, so an automated message still has to pattern-break to land, and until that is solved AI SDRs may be "a temporary dead end" [5]. More fundamentally, you are selling to people and you have to stand out, and AI can be creative in instances but cannot continuously do it [4]. He concedes the whole debate turns on whether models improve by one or two orders of magnitude, and says he has no strong intuition about how close that is [2][5].

On data as a commodity and building for wrongness

The industry assumption he inverted early was that data is the moat. His view is the opposite: "Data is a commodity," get it from wherever it is needed, which is why integrations were first-class citizens from the beginning, each living in its own AWS Lambda function so junior engineers could ship them without risking the whole product [3]. Clay aggregates every provider rather than betting on one, since which data you need depends on what you are looking for [2][5], and treats providers as collaborators: customers can spend Clay credits, or bring their own API key for free, and once a customer is large enough he would rather they contract directly with the provider, because the value Clay captures is in what you do with the data, not in reselling it [2]. LLMs then produce properties no provider sells, such as whether a company has a remote office in South America or mentions AI in its support documentation [2][3].

The second inversion is accuracy. Everyone else claims their data is complete and best and only, "and everybody knows that's a lie" [3]. Clay starts from the assumption that "the data is going to be wrong and messy. In fact, we're not even responsible for it" [3], and builds accordingly: session replay was a P0 so you can watch how Claygent got an answer and where it went wrong, agents that check other agents, flags on likely errors, fast manual review, a playground with prompt versioning [3]. He goes further and argues there is no canonical dataset for companies and people, because they are living entities seen from different perspectives, and the ability to ask any question about them is the edge [3]. The philosophical position underneath is a bet about timing: some competitors are betting that two orders of magnitude of model improvement will make all of this unnecessary, and he says Clay does not need to win that debate, only to ask "what is working right now" and earn the momentum to build what is next [3]. At scale the hard problems are rate limits measured in tokens rather than requests, customer-supplied API keys used elsewhere, and agents that fail at step three of a seven-step plan, where the common complaint is simply not knowing what is happening or where it broke [3].

On outbound that deserves a reply

Confronted with the objection that Clay arms one side of an arms race while filtering tools arm the other, he invokes the Red Queen and then refuses the premise [2]. Clay is not trying to help people send more messages: "you should be doing deeper research ahead of time," pick smaller deeply researched segments, and send a more worthwhile message, at a scale that only becomes feasible with tooling [2]. This is why the company stays out of the messaging channel, connecting to third parties rather than owning it, since people use Clay for mail, gifts, websites and ads and Clay is agnostic about the endpoint [2]. The discipline is enforced by consequences: "if you don't use us correctly you're the one who's going to get banned not us" [2]. He also declines the standard growth promise, on the grounds that if your product is weak your message will not work regardless of targeting, and "what you're selling might not be worth growing" [2]. That negative result is still information, and it is what makes Clay aligned with recipients rather than against them [2]. He frames the underlying market simply: there are only three ways to grow, find new customers, convert them, or expand existing ones, and Clay serves all three [2], including expansion plays like telling a customer's head of engineering that their designers are already heavy users [2].

On what comes next in the product

The arc he describes runs from best-in-world enrichment by aggregating providers in a spreadsheet, to first-party signals such as repeat website visits, to third-party signals that change over time, including monitoring competitors' sites for messaging changes as an indicator of what is working for them [3]. Audiences makes companies and people first-class objects so signals can accumulate on them [3]. A custom sequencer exists because existing sending tools were built for substituting pre-populated strings rather than AI-composed messages, which cannot be entirely model-written and need specific spot-checking [3]. Sculptor was built as an agent that helps you build in Clay, aimed at the user who has an idea but does not know the tool, with the further step being the user who knows neither: connect the CRM, identify which properties predict good customers, find lookalikes, mine what worked on previous customers, and send or brief the team [3]. He notes Sculptor's most powerful use turned out to be something else, asking questions about tables you have already built [3]. He describes the same pre-configuration idea elsewhere: read your site, infer your targets, set up the searches and data, ingest your wiki or sent emails to match your language, then let you tweak, which sounds automated but is really Clay assembling its building blocks with you [4][5]. The near-term prize he sees is stitching scoring, research and messaging into one end-to-end campaign loop that reports whether it worked [5], and beyond twelve months he says he is staying open-minded [5].

On risk, courage and long-term greed

Asked what statues he would build, he starts with courage, because "capitalism rewards risk" more than hard work or skill, a claim he supports by pointing out how many extremely hard-working or extremely skilled people are not rewarded [1]. Real risk has two tests: "you need to not know what's going to happen genuinely," and it must carry a high potential for shame, because that is when you are about to discover something new [1]. Plenty of founders think they are taking risk and are not, and his example is the founder serving both sales and recruiting, to whom he said "You have to pick one," since value comes from committing to a specific group of people in a specific way [1]. He applies the same standard to Clay's own history: the takeoff came less from technology than from the courage to commit to non-obvious decisions and follow them to their logical conclusion [1].

His second statue is justice, argued from self-interest rather than sentiment. He wants a stable society with time to think, stability requires justice, and no arrangement where one group dominates another is stable, because the mistreated party is a tremendous nuisance and you cannot bluster your way out [1]. He translates that into treating people with respect and fairness in every situation, hiring, firing or disagreeing [1], and connects it to democratization: people are stable when they believe they have a fair chance at what you have [1]. The personal version is a soccer instinct, that winning after a foul you got away with does not feel as good, which he reads as evidence he is "a long-term greedy" person, a phrase the company uses, and his argument that staying in integrity is the only way to be long-term greedy [1]. On motivation he is unsentimental: chasing adulation, prestige or money is empty and passes, company narratives rarely match founders' real motives, and the only durable measure is your own self-respect, which for an ambitious person is a higher bar than anyone else's because you know everything you have thought and done [1]. Becoming more courageous, he says, starts with being honest with yourself about why you are doing something [1].

On running the company

He argues that a fast-growing company should be reducible to a few assumptions from which most decisions follow. Clay's three were: people in go-to-market are creative, so give them the most powerful tool; that requires the right user, so name and serve the RevOps-adjacent go-to-market engineer; and charge for usage rather than per seat, so that customers doing more with fewer people is not an anti-incentive [1]. If those hold, anyone can work out what to do next, and communicating three constraints beats issuing instructions [1]. Alongside that he gives the team a time budget: "90% of your time should be executing 10% meta reflecting on anything otherwise you're wasting your time," which is how he handles speculation about where models go, on the grounds that whatever the future options, the more successful Clay is now the better positioned it is to choose [2].

Culturally he describes "chill High Achievers": anxiety, fear and frustration can be acknowledged without being reacted to in the moment, because clarity of mind is what lets you see all the pieces and then commit [4]. Interviews follow the same rule, collect and share information rather than judge in the moment, then decide at the end [4]. He is comfortable running the company on held tension, invoking Hegel's thesis, antithesis and synthesis: creativity against automation, physics against making music, a canvas that also has to run at volume [4].

On AI and jobs

He resists the headcount-collapse story, reaching for induced demand: when you build more roads you get more cars, and the system is set up to convert efficiency gains into more growth rather than the same growth with fewer people [2]. Some companies will choose the small, high-revenue-per-employee path and be satisfied, but in winner-take-all markets competition at the top pushes everyone to do more, so he expects more productive people and new areas of work that were not worthwhile before, rather than fewer people [2]. He does not think human sales reps go away soon [4], and expects SDRs to remain, doing follow-ups, calls and the parts that are not yet automatable [5]. The important caveat he adds is that most predictions of job loss assume one to two orders of magnitude of model improvement, and if that arrives many jobs do become less needed [2]. What he does think is disappearing is the old craft of stitching together tools and grinding out enriched spreadsheets, which belongs to automation; the new work is figuring out what actually predicts a good customer [4].

Takeaways

• The founding ambition was never sales software: it was giving the power of programming to an order of magnitude more people, with the spreadsheet chosen as "the world's most popular programming environment" and then connected to the internet's data [1][2].
• Treat go-to-market as creative, not procedural. The purpose is go-to-market alpha, because scaling any tactic raises the baseline and kills it, so what worked must keep changing [1][3].
• Data is not a moat. Build integrations as first-class citizens, assume "the data is going to be wrong and messy," and ship trust features like session replay as P0 rather than promising accuracy you cannot deliver [3].
• Reject both the copilot and the fully autonomous AI SDR in favour of a third way: centralize the work with a technical GTM engineer who automates what is automatable [4][5].
• LLMs are already strong at account research, similarity-based scoring, summarising calls and emails, and generating snippets and value props; end-to-end autonomy fails because plans commoditize and any broken step is a disaster [5].
• Send fewer, better-researched messages; refuse to own the sending channel; and tell customers plainly that if the product is weak, no targeting will save the campaign [2].
• Run a fast-growing company off a small set of stated assumptions, including usage-based pricing so that customer productivity is not an anti-incentive, so that anyone can derive the next decision [1].
• Real risk means genuinely not knowing the outcome and being able to fail shamefully; commit to one specific customer rather than hedging across two, and measure the result by self-respect rather than prestige [1].