Steve Cox runs a software company of about a thousand people, put together from two businesses that merged at the end of last year, and he says AI-native startups now come to him at a rate of at least one a day.
The consensus for two years has been that companies built on AI from scratch would take the market from the traditional software vendors. Cox's argument is that the vendors are absorbing the technology instead, and that the startups selling a single task are the ones running out of road.
"I get at least one a day that raise some funding, sold to some customers, aren't getting traction, adoption and usage seems to be the biggest challenge."
Cox was appointed chief executive when Clari and Salesloft combined, joining two weeks after the deal closed, and his first job was to hire an entirely new executive team; he has spent more than 25 years in software, most recently running Employ.
The full interview is covered here so you can skip it. 25 minutes of audio, 15 minutes of reading.
Here are the 12 arguments that matter.
👤 Guest: Steve Cox, CEO of Clari + Salesloft, the revenue software company formed by the merger of Clari and Salesloft, previously CEO of Employ
🎙️ Host: Firas Sozan, Founder of the cloud, data and AI recruiting firm Harrison Clarke, who runs Inside The Silicon Mind
📰 Published: 15 September 2026 on YouTube (Inside the Silicon Mind with Firas Sozan)
🔴 YouTube | 🟣 Apple Podcasts | 🔗 Episode page | ⏱️ 25 min | ✅ Time saved: 10 min
Key Takeaways
The AI-native startups arriving on his desk are stalling at retention, not at the sale
They reach a level of recurring revenue, cannot keep customers, and then cannot raise again
He expects the phrase "AI company" to disappear within a few years
Not because the technology fades, but because every software vendor will have it
The SaaS-is-dead selloff hit companies that were still generating free cash flow
Owning one task in a workflow is the thing that does not survive the shakeout
Most acquisitions fail on culture and incentives rather than on the technology
He tells acquired employees directly that they never signed up for the company they now work for
Building agents you can build is not the same as building agents worth building
He says most people he talks to are now spending more time in front of a computer, not less
He runs the merged company on three numbers and three constituencies, and nothing else
1. SaaS Isn't Dead, It Evolved
Sozan opened on the consensus view — that AI-native startups would replace traditional software companies — and asked what Cox makes of the SaaS-is-dead narrative.
Cox put the moment in a sequence rather than treating it as a break. Software digitized paper, then the internet era ran its course, then on-premise systems moved to the cloud and the cloud to subscription software. What is happening now, in his framing, is the next transition in that same line.
The new layer is agentic software sitting inside existing platforms, not replacing them. He described it as "the next phase of transition into really which is largely an agentic capability which is software that sits inside platforms like ours and others", which he said will let people work more efficiently alongside it.
He was asked the question often enough that he flagged it as one he gets asked a lot and thinks about a lot.
2. The SaaS Apocalypse Trade
Sozan pressed on why the narrative got so loud over the past year or two. Cox answered as a shareholder as much as an operator.
What he found interesting was the share prices. Companies that were largely profitable and generating a lot of free cash flow got depressed valuations on a perceived lack of opportunity, or on the view that AI was coming to take the business.
He said there is a lot of money and a lot of capital ready to be invested, and AI is the next wave of people trying to capture it. He pointed to the large valuations attached to the leading AI labs as the sort of thing investors want a piece of early.
On the industry's own habit: "I think in this industry and the tech industry we're very good at selling the hype of new technology up front" — with the real value and the delivery arriving later.
What he says he is watching now is the swing back: traditional software vendors being assessed on the AI capability they have built and what they bring to the table, so buyers can find more rounded solutions.
3. From AI FOMO to ROI
Asked what signals tell him customers have moved from experimenting with AI to demanding business outcomes, Cox answered from the volume of conversations he has with chief revenue officers, finance chiefs and chief executives.
The first phase was fear. "I think there was this whole FOMO around AI where it was like if you don't do AI then your business is going to be dead in a year." People felt compelled to go and find out, and "what is your AI capability" became the new question.
Buyers went looking for AI-native point solutions because that looked like the fastest route in. What they found, he said, is that it is not as easy as bringing in eight or ten different point solutions to solve an end-to-end business problem, managing them, and then keeping them growing in features and value.
His summary of the result is blunt: "A lot of AI projects don't deliver the ROI people expected." He credited the published statistics rather than his own count.
The conversation he now has repeatedly is a different one — how to embed AI capability into the business to drive real return and give time back, rather than how to acquire AI.
4. Who Survives the Shakeout
Sozan asked what separates the AI companies that become the next Snowflake or ServiceNow from the thousands that disappear.
Cox's answer is scope of workflow, not quality of model. Everyone has to start somewhere, and a tool that solves one specific problem is a legitimate start.
The test is whether the company can grow out of it: "I think what drives success is how do you start to break out of that specific problem you solve and solve bigger chunks of the workflow", adding more value to the business overall and giving meaningful, tangible return.
He put his own company on the other side of that line, describing a platform covering the end-to-end funnel, its lifecycle, engagement, intelligence and forecasting.
5. Human in the Loop
Asked directly whether AI will replace humans, Cox allowed the narrow case and rejected the broad one.
"I'm definitely a subscriber to human in the loop." There is a lot AI can do to take away mundane, repetitive tasks and give time back to spend with customers and partners.
Reliability is what stops it going further. He cited hallucination, and customers who ran agentic queries, got results back they did not trust, and were given a completely different answer when they pushed back.
His timing call is explicit: "I think we've got work to do to get to a level of viability before we fully replace humans and I don't think it's something that we see in short order."
The underlying reason he gave is that the systems are not deterministic — the same request phrased differently produces a very different response — and that the output depends on the data fed in.
6. Just Because You Can Build
The exchange turned to whether the productivity gain survives the checking, in a question from Sozan that named the problem precisely.
Sozan's own worry: "But how much time is actually spent using AI, but then having to proofread what AI has done to make sure there aren't any bugs and if it's a document, it hasn't made a mistake."
Cox reached for the buy-versus-build debate he says is live with his customers: "I have this philosophy of like just because you can build it doesn't mean you should." The quality and accuracy of the data being trained on is what decides whether building is worth it.
He went further than caution: "I wonder like how many people are building agentic capability that's really a fool's errand." Most people he talks to are spending more time working, and more time in front of a computer, because they now have agents they can prompt into building things they could not build before.
The question he says has not been answered is whether those things are relevant or add value. He is clear that plenty of genuine use cases exist — engineering in particular is seeing a large uptick in productivity, and putting the work in up front does pay back.
What is missing, in his account, is the planning and the architecture: separating what really gives time back from busy work nobody could do before and nobody needed.
7. Just Software Companies
Out of that came the prediction the episode turns on.
Cox accepts that some people will prompt, vibe-code and build systems they can run a business on, but says the mainstream will get its AI from software vendors, as with any other technology trend.
He expects genuine winners born as AI companies — companies that grow, take market share and do not exist today. He said there will be a lot of those.
His central claim is that the incumbents make the journey too, and the category dissolves: "We'll just be talking about software companies." The prediction is for a few years out.
8. The Consolidation Wave
Sozan moved to the second topic, and quoted Cox back to himself from their preparation call: a major wave of consolidation is coming in AI. He asked what Cox sees that makes him think hundreds or thousands of AI startups get acquired rather than becoming standalone businesses.
The entry cost is the starting condition. Because it is easy to get going with relatively few dollars, thousands of entrants are coming in and many are attacking the same specific problems — in his market, the intelligence, engagement and forecasting pieces of go-to-market.
The barrier is not the model, it is the archive: "the problem is there's only a handful of companies that sit on a lot of historical data that they can use to train models to get real good accuracy and context."
The pitch deck arrives daily. "I get at least one a day that raise some funding, sold to some customers, aren't getting traction, adoption and usage seems to be the biggest challenge." The cause he names is that these products ask people to work differently in order to solve one problem.
"These are not fully built out platforms that can solve end to end workflows for people."
The failure sequence he described runs through retention rather than sales: a company reaches a level of annual recurring revenue, struggles to keep customers, cannot get to growth as a result, and stalls at the next funding round.
Those companies are now coming to market looking for acqui-hire deals or equity deals, and he was direct about what they are worth: not high-value, cash-rich businesses.
9. Why Most Acquisitions Fail
Sozan said change management in service industries can be brutal, then asked why some acquisitions work and the majority do not — the acquired company becomes a feature, the team is absorbed, and the brand disappears.
Cox said there could be a hundred reasons, and that it starts with why the company was bought at all — a defensive move, a technology addition, or a customer base. The rationale changes everything that follows.
His main answer is culture, and specifically that startup culture cannot be recreated inside a larger company. He described a 15- or 20-person team in one location building a strong, family-like culture on long hours, pizza, parties and visible growth, and said that is hard to maintain at scale — harder still in a remote world.
Incentives are the second half. The programs that draw people to startups are not always available inside a larger organization, and they come with risk.
The mismatch is about pace, not money. Someone who likes the startup buzz and the not knowing whether they are building a unicorn can find themselves in a company that is more stable, more organized and on a different pace, and realize it is not for them.
His criticism of how acquirers behave is about listening: "And I think where people do a bad job in M&A is actually understanding what the people want from the company that come in."
His own practice is to name it out loud. He tells acquired employees they never signed up for this, that they joined one company and now work for a completely different one, and "So first thing you need to ask yourself is this somewhere I want to be?"
The answer depends on the thesis behind the deal: whether the business is run standalone, whether the founder stays or goes, whether it is absorbed into a sales unit. Each changes the culture dramatically.
10. Building a Company to Sell
Asked how a founder of an AI startup today should think about being acquired, Cox split the question in two.
"I think if you're building a company to exit, then you need to know who you're going to exit to." Knowing why you are building the company comes first; defining the exit strategy follows from it.
The second half is deliberate relationship building — engineering the partnerships and conversations that give you the best set of options.
His worked example is Salesforce, which he said has acquired a number of partner companies that complement its technology stack and fill gaps. He does not know how many of those were built with that outcome in mind.
The fork he draws is simple: either you are building it to sell, or you are building it for high growth — and if you have high growth and can keep raising money, he asks why you would sell.
He closed the topic on the upside case, and kept his own hedges on the numbers: "I think YouTube's revenue is now something like a few billion per week in comparison how much they were acquired for which I think was a billion or a billion and a half, something small like that." He named Instagram alongside it as an incredible acquisition story, and called this a very exciting time to be in technology.
11. Merging Clari, Salesloft
Sozan asked him to unpack where he is today and what the job actually is. Cox dated the merger to late November and said he joined in December, two weeks after close.
"My first job was to hire out an executive team." His reasoning: bringing together two companies of equal size creates a company of a different scale that will grow at a different pace, and that sometimes needs different talent. He said the team is pretty much in place.
The second job was listening to customers on both sides — what they like about the technology and what they want to see — because that informs the road map.
He said AI makes the integration itself easier, which he contrasted with his own past experience. Merging two technology sets into a single user experience has always been hard; now he can build above the data layers, create a single context engine to pull the data together, and surface it anywhere in the application. Embedding one application inside another has also become easier.
He pointed to a July release bringing the two products together with new functionality.
His evidence that it is working is one customer: an ex-Clari company new to Salesloft, already seeing the benefit of the combination and specifically the agentic research and coaching features, which he said help managers understand what their team is doing, get the team to the right customers, and reach a more predictable forecast.
He then asked the definitional question back at himself: "At what point do we become an AI company, I mean like is it one agent, is it two agents, is it 10 agents?" His own view is that the company has a lot of AI capability now, that it is market-leading in the category, and that it obviously did not start there.
12. Three Things He Tracks
The show asks every guest for a book recommendation, and Cox's answer turned into the closing section on how he runs the company.
His pick is Ben Hunt-Davis's Will It Make the Boat Go Faster? The concept, as he put it, is that "every decision that you make, you should pressure test against the mantra of will it make the boat go faster" — the rule the British rowing eight used to win Olympic gold for the first time.
His objection to how focus is usually practiced is that people claim it and then break it: everybody talks about focus, "and then they're focusing on 25 different things". His conclusion: "So, they don't focus at all."
Sozan asked how big the organization is. About a thousand people, across the two merged companies.
Cox's three are product innovation, bookings and revenue, and retention, and he asks himself daily whether what he is doing drives results in those areas.
He triangulates every decision against three constituencies — customers, investors and his own people — and checks that the decisions on those three priorities are aligned to all three groups.
Running across everything is what he called identification: how the company identifies its product, its internal processes, and how to make the best use of technology.
The structural advantage he claims is dogfooding: "we are customer zero for everything we do." The company sells software and has to use it every day to sell, which he says produces a lot of good feedback on the product.
His own restatement of the three: "It's like product innovation, churn, like retention, and then, bookings, revenue, with an AI thread run all the way through."
Bonus Insights
Cox's closing view is that the SaaS-is-dead narrative created an opening in the stock market for him personally.
"Some of these companies are printing cash and sitting on lots of cash as well." He expects many more large acquisitions through this round of transition, and said M&A could help round out parts of his own product plan.
Sozan disclosed that he ran an episode a month or two earlier arguing the opposite case — that SaaS is dead — and said the contrast is the nature of a market moving this fast.
Cox said engineering is the function where he sees the clearest productivity uplift from AI today, and that he believes there are plenty of other use cases still to be identified.
Sozan floated a rerun of the conversation in person, on the subject of merging two companies into one software business.
Cox's bottom line is that the AI shakeout is a distribution story rather than a technology one: the startups that own one task are running out of retention and coming to market, and the vendors that already hold the historical data and the end-to-end workflow will end up carrying the capability — at which point nobody calls it an AI company any more.
Products, Companies & Tools Mentioned
Clari and Salesloft (The two companies Cox merged, now about a thousand people; the July release brings the two products into one)
Salesforce (His example of an acquirer buying complementary partner companies to fill gaps in its technology stack)
Snowflake and ServiceNow (The archetypes Sozan used for AI companies that make it to scale)
Instagram and YouTube (The acquisition stories he calls incredible, with his own hedged figures on what YouTube now earns against what it cost)
Employ (The company he ran before taking over the merged business)
Books & Resources Mentioned
Will It Make the Boat Go Faster? – Ben Hunt-Davis (His book recommendation and his decision rule: pressure-test every decision against whether it makes the boat go faster)
Watch the full episode:
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: