AI to AE's: Grit, Glean, and Kleiner Perkins' next Enterprise AI hit — Joubin Mirzadegan, Roadrunner
Summary
Roadrunner is Joubin Mirzadegan’s bet that AI-era pricing is breaking legacy CPQ’s underlying data model. Simple per-seat contracts have exploded into 30, more than 50, and in some cases 100 products, volume discounts, early renewals, expansions across, say, 15 product lines, and consumption billing. “These pricing models have gone bananas,” while sellers remain trapped in “loading-screen hell.”
The market signal came from two separate groups of five technology CIOs, six months apart, unanimously naming CPQ as their top problem. Joubin’s broader network spans 35 CIOs from companies including Uber and Box; KP mapped the market but found no compelling startup. Salesforce, which he says has 95% share, has ended its existing CPQ product, creating what Roadrunner sees as a two-year race against its replacement.
Roadrunner’s prospective moat is founder-market fit plus unusually direct enterprise distribution. Episodes 1–80 of Joubin’s podcast connected him with CROs feeling the pain, while KP’s CIO network connected him with those responsible for fixing it. Four deliberately “hairy” design partners now pressure-test every SKU and hardware, software, SaaS, consumption, channel, discount, and other permutation; customer demand exceeds engineering bandwidth.
Glean taught Joubin that an enterprise category dismissed as dead can still produce a major AI company—but only after solving permissions, deployment, and trust. Arvind Jain built initially for Rubrik, giving Glean an accessible design partner and proof that enterprise search worked in production. Selling it still resembled “passing a bill through Congress,” with technical stakeholders, security, champions, and resistance all requiring alignment.
Windsurf’s reported rise from roughly $0 to $100 million ARR in seven or eight months came from treating distribution as seriously as engineering. Joubin’s formulation was “Google-class product and Salesforce-class distribution,” reinforced by market pull, fast hiring, and rapid execution. Its go-to-market team’s move to Cognition underpins the host’s thesis that strong Devin product and engineering can now be paired with an unusually effective coding-sales machine.
Early sales hiring should optimize for potential, technical curiosity, and stage fit—not prestigious logos. Across five executive functions at KP’s top eight companies, Joubin said 38 of 40 leaders were reporting directly to a CEO for the first time. A fighter who sold the number-three product may outperform someone who inherited brand, inbound, and a playbook at $100 million ARR; early selling is “more like art than it does science.”
The conversation rejected inflated startup missions and framed grit around passion rather than permanent suffering. Roadrunner must first solve one painful workflow, delight users, and then “earn the right” to tackle adjacent problems and build a compound company. The host cited Angela Duckworth’s definition as passion plus sustained perseverance; Joubin said passion was the part he valued most, because genuine care makes the work feel light and helps him outlast others without continually gritting his teeth.
Deep dive
1. Joubin built a podcast to create the sales network he lacked
Joubin entered KP after a startup sales career and a run building Palo Alto Networks’ central-US public-cloud business from zero. His initial charter was to manage a CIO network and help technical founders answer the post-product question: “What do you do next?”—especially around sales and distribution.
Working with Arvind Jain on Glean exposed the gap. Joubin and Arvind could design go-to-market “routes,” but neither could run them, and Joubin lacked the 30-year coaching tree needed to recruit the right sales leader. The podcast became “an excuse to get to know people that I do not know today.”
KP’s initial reaction was more skeptical than supportive, so Joubin recorded an episode with his former boss and sent it around: “This is what it would sound like—and I still have a job here.” The firm agreed to try 10; Joubin committed to reaching 100 before judging its quality.
That commitment insulated him from the vulnerability of publishing and from “the comments,” where reacting to every opinion can kill a show before episode five. Moving from CROs to founders also changed the conversation: founders can speak with authority without constantly wondering, “What is my boss going to think?”
2. Glean succeeded inside a category buyers had trained themselves to reject
When Arvind pitched enterprise search around 2018 or 2019, the category produced an industry-wide eye roll. CIOs and investors had heard decades of promises—including from Google—that somebody would finally solve it. Joubin’s retrospective is important: today Glean sits “in the heart of the hurricane,” but its outcome was “extremely unobvious” then.
The technical burden went far beyond search quality. Glean had to crawl an organization’s entire corpus while preserving every permission and authentication layer, ensuring, for example, that Joubin could never see swyx’s compensation. Only after that came deployment architecture, security reviews, and the sensitivity of indexing a company’s internal information.
Arvind’s advantage was Rubrik, where he had encountered the problem and which became Glean’s core design partner. He knew the systems and people, had access for co-development, and could show skeptical buyers that the product already worked in production.
Joubin compared enterprise adoption to “passing a bill through Congress”: the founder must map every stakeholder, explain what is in the product, align supporters, and help the customer manage its own process. The host sharpened the analogy into a military campaign—identify champions, locate resistance, and prosecute the rollout deliberately.
3. Enterprise AI customers must learn the platform and the product simultaneously
The host argued that enterprise AI can be harder for engineers now because customers face two learning curves. They must decide how LLMs fit their organization while adopting a specific application built on top of a shifting foundation. Joubin agreed that customers are first figuring out the underlying stack and then trying to use a product on top of “quicksand.”
That explains the rise of forward-deployed engineers. The vendor first teaches the customer where LLMs work, then embeds an engineer to co-develop a solution fitted to the customer’s environment. At this stage of AI adoption, extensive technical handholding is not an exception; in Joubin’s view, it is the natural go-to-market motion.
The broader lesson from Glean is that good technology does not dissolve enterprise friction. Permissioning, security, deployment, stakeholder politics, and workflow design remain load-bearing—and early believers willing to co-develop can matter as much as the initial model capability.
4. Windsurf paired product ambition with a first-class sales machine
KP expanded Joubin’s original charter into a platform for technical founders: Liam helped deepen sales support, the team increased access to world-class customers, and Suzanne helped with demand generation and marketing. The premise was that exceptional product engineers often have never closed a deal or built a top-of-funnel machine.
Windsurf became the clearest specimen. Joubin described “something like” $0 to $100 million ARR in seven or eight months as the most torrential growth he had seen in a KP company. The founders committed from the outset to both “Google-class product and Salesforce-class distribution.”
Three mechanisms carried that growth: coding was being pulled aggressively by the market; expensive engineers benefited from AI copilots that could reason over structured code; and the company moved, hired, and operationalized quickly. Joubin repeatedly resisted understating the distribution piece: Windsurf was as serious about sales as it was about product.
The host’s physical evidence was an entire office floor dedicated to video production—unusual for a young developer-tools company, but consistent with its distribution priority. With that go-to-market team now at Cognition, he sees a potent pairing with Cognition’s Devin product and engineering.
5. Great startup sellers are discovered through motivation and context, not logos
A brief career-ranking exchange put people first—Joubin would “double people”—then market and product, with money later. His recruiting corollary: do not see Snowflake or Databricks on LinkedIn and assume the person can build sales for an early AI startup.
Someone who fought “tooth and nail” selling the number-three product may carry more useful ability than a seller who joined Snowflake around $100 million ARR and inherited its Bay Area enterprise brand. The latter may fit that scaled context perfectly; the mistake is importing them into a startup that has none of the same machinery.
Joubin’s portfolio evidence was striking: across five executive roles at KP’s top eight companies, including Rippling and Glean, 38 of 40 leaders reported to the CEO for the first time. The hiring signal was not prior title repetition, but learning speed, trust, accumulated context, and the internal flame that survives repeated rejection.
AI also raises the technical bar. Sellers need not explain every transformer detail, but they should question engineers deeply enough to articulate the product and its ecosystem without outsourcing understanding to a sales engineer. Willingness to get into the product is itself visible in a candidate’s history.
6. Sales leadership changes from improvisational art to repeatable science
Stage similarity matters as much as capability. A seed or Series A company should be wary of sellers who previously joined companies only after those companies had $50 million or more in ARR: they have always had brand credibility, inbound leads, and an established playbook to execute.
The first sales leader or AE instead operates “way more as an artist” than a scientist, finding creative ways to make an undefined motion work. At Windsurf’s later stage, that creativity had become a machine of boot camps, battle cards, qualification criteria, and repeatable execution.
The host supplied a useful counterpoint from Netlify: when engineering worried about missing competitive features, a new sales leader replied, “Give me anything, I’ll sell it.” That old-school confidence remains valuable, but the conversation’s conclusion was that AI selling increasingly requires genuine product understanding alongside persuasion.
Joubin even doubted he would fit Windsurf’s scaled motion today; meticulously executing a handed-down qualification playbook “is not my thing.” The admission reinforces his central hiring point: performance depends on matching the person to the company’s exact operating phase.
7. AI pricing complexity has pushed legacy CPQ beyond its original design
Roadrunner began with Joubin’s recurring sales nightmare: a 30-second page load while trying to close a deal with two days left in the quarter. Teams were blamed for demanding one- or two-day turnarounds, although the underlying software could not move at revenue speed.
Legacy CPQ assumed a static mapping between users and licenses—say, 1,000 LinkedIn seats. Modern vendors may carry 30, more than 50, or in some cases 100 products, with volume tiers, discounts, renewals, early renewals, expansions, channels, and transactions spanning, say, 15 product lines.
Consumption compounds the problem. Customers may commit to a minimum and then pay for excess usage, while AI vendors increasingly price around consumed tokens. Joubin believes AI-era pricing models will, “at a minimum,” start to look more like consumption-based pricing, meaning the pricing explosion has “barely even started.”
The external validation was unusually uniform: two different dinners, each with five CIOs and separated by six months, produced the same number-one problem—CPQ. “Pain does not grow on trees like that.” KP mapped the market and found nothing compelling; GPT-3.5 then supplied the missing technical catalyst.
8. Roadrunner has a two-year window and a deliberately adversarial design process
Joubin’s light-bulb insight was that LLMs can reason across structured and unstructured text related to pricing rules, much as coding systems reason over code or Harvey reasons over case law. Enterprise workflows and controls can then be layered on top.
Incumbents face an architectural problem, not a missing feature. Joubin compared it with companies initially lifting applications from on-premises infrastructure into AWS, then discovering they needed cloud-native redesign. Supporting consumption, SKU sprawl, and interconnected rules similarly requires rebuilding CPQ’s data model “from the ground up.”
He said Salesforce has 95% market share, has ended its existing CPQ solution, and is making customers move toward a replacement he considers nonexistent or “incredibly flimsy” today. Roadrunner therefore sees a two-year opportunity to outrun a roughly 100,000-person incumbent—a race Joubin prefers to competing with OpenAI.
Distribution completes the timing thesis. Podcast episodes 1–80 connected Joubin with CROs suffering the problem; KP’s network connected him with CIOs delivering the software, often alongside CEOs. He called that “unfair distribution,” and concluded that hiring an outside founder would make less sense than building the company himself.
9. Roadrunner must solve the hardest quotes before earning a broader mission
Roadrunner had nine people at recording. Technical co-founders AJ and Eugene met at Caltech; AJ entered at 15 and finished second in his class. AJ went to Robinhood, Eugene went to Meta, and they built Mars-rover software together at NASA. They later created Athena, an LLM application for college admissions, and had independently encountered CPQ pain through conversations with founder friends.
The company is co-developing with four design partners through a shared Slack channel and weekly stand-ups. It chose the “hairiest” customers—hardware, software, SaaS, consumption, every SKU and rule—so an unforeseen permutation will expose the data model now rather than after deployment.
Those rules can connect geography, seller permissions, products, maximum discounts, and channel structure: a UK AE may quote only certain SKUs, while a Canadian channel transaction triggers another rule set. The team spent a large portion of its time on the data model alone, throwing real-world complexity at it until it stopped tipping over.
The intended payoff is a system that uses historical quotes to recommend packaging and delivery. A Costco opportunity might resemble a prior Nordstrom deal. Today that judgment requires calls to finance, the deal desk, and top AEs; Joubin frames Roadrunner as automating deal-desk administration and augmenting AEs so those people can spend more time on strategic work.
10. Constraint, routine, and genuine enthusiasm are Joubin’s operating system
Roadrunner is not demand-constrained: Joubin said customers are “banging down my door.” His recurring disagreement with his co-founders is that he wants to admit another 10 while they insist engineering cannot support them; the roadmap is already “being dragged out of us.”
In the mission discussion, the host argued that Roadrunner solves a hard problem for a specific set of high-value users rather than addressing mental health or hunger. Joubin agreed with the principle of solving one problem, delighting users, and then “earning the right” to solve the next, while acknowledging that founders sometimes have to pump themselves up for a fundraise.
His personal system removes recurring decisions: exercise and sweat every morning, bike Hawk Hill on Mondays, run twice and lift twice weekly, play a sport, and eat a salad for lunch. “It is way easier to know that I am going to work out every day” than continually negotiate a 90% commitment.
The host cited Angela Duckworth’s definition of grit as passion plus perseverance over a sustained period of time. Joubin said passion was the part he valued most: genuine care makes effort feel light rather than requiring constant teeth-gritting. His favorite definition of sales is “the ability to transfer enthusiasm from one person to another.”