This interview covers how Ramp uses AI to automate corporate finance, aiming for a 'self-driving' expense system. Founder Karim Atiyeh believes technical founders have an edge in the AI era because they can see possibilities others miss—like Ford building cars when people only asked for faster horses. Key mentions: Amex (strong brand, weak product; Ramp aimed to match it), Concur (poor user experience, used as a cautionary tale), and Capital One (acquired Karim's previous startup Paribus, rebranded as Capital One Shopping).
Ramp co-founder and CTO Karim Atiyeh explained on the program how the company is building an "autopilot finance" system through AI, becoming the fastest-growing financial automation platform in history, with revenue surpassing $1 billion within five years. The core thesis is: transforming AI from a
Karim Atiyeh is the co-founder and CTO of Ramp, the fastest-growing financial automation platform in history, surpassing $1 billion in revenue within five years. The core theme of this episode is how Ramp is building an "autopilot finance" system by transforming AI from a productivity tool into the product itself, eliminating bureaucratic waste in enterprises. Karim Atiyeh believes that technical founders will dominate the AI era because they can see possibilities others overlook—just as people only asked for faster horses before Ford built the car.
Karim Atiyeh argues that most companies remain stuck in the first phase of using LLMs to assist existing tasks, while the real transformation lies in programming the LLM as the product itself.
Atiyeh divides AI adoption into two phases:
Ramp's policy agents are a quintessential application of this concept. Traditional enterprise expense management relies on a static travel reimbursement policy document, where employees self-check before transactions and finance staff manually audit afterward, resulting in extensive back-and-forth communication. Ramp's policy agents, however:
"You're not using deterministic code to tell the software which transactions comply with the policy. Instead, you have a living, evolving text document that instructs an agent with access to tools like calendars and emails on how to classify transactions." (Karim Atiyeh)
The same principle can be extended to invoice processing: verifying whether an invoice is fraudulent, whether it corresponds to goods ordered and received, and whether the price matches what was negotiated — most of these processes are undocumented, but Ramp infers policies by observing a large volume of customer behavior, driving the next generation of agents.
Karim Atiyeh argues that the previous generation of enterprise software served only decision-makers, leaving other employees to endure "death by a thousand paper cuts" every day. Ramp, by contrast, pursues an Instagram-level user experience.
Atiyeh points out that the typical model for traditional enterprise software (e.g., Concur) is: build for decision-makers → convince them to switch → stop caring about user experience after the switch. Ramp's approach is:
"We always ask ourselves: How can I find the answer to this question on my own, without having to ask the customer? Or, if I'm telling the customer to do X, why can't I do it for them?" (Karim Atiyeh)
Extension of the design philosophy — brand differentiation: Ramp chose yellow as its brand color because all financial apps fall into the blue or green spectrum (green = money, blue = trust), while yellow is only used by Snapchat. Atiyeh believes that seeking differentiation is itself a strategy — "If you do it well enough, it will pay off over the long term."
Karim Atiyeh places "divine discontent" at the core of management, arguing that celebrating lagging indicators undermines the urgency needed to solve current problems.
Atiyeh cites Jeff Bezos’ philosophy: the current quarter’s results are determined by work done 6–12 months ago, making it odd to celebrate today’s numbers. This culture manifests in:
Speed in practice: Atiyeh opposes detailed effort estimation before development begins, arguing it incentivizes teams to provide longer estimates to ensure they hit targets. "You can know exactly how long something will take, or you can do it very fast — it’s hard to have both." (Karim Atiyeh)
Lessons from the Paribus era: As the sole engineer, Atiyeh was forced to learn delegation and management after breaking his arm in a skiing accident. This made him realize that maximizing the team’s total output is more important than his own direct output.
Karim Atiyeh explains why Ramp shifted from a pure transaction volume (TPV) model to a hybrid SaaS model: card transaction volume tends to plateau as enterprise complexity grows, but software value increases with enterprise scale.
Atiyeh notes that small businesses route nearly all spending through cards, making the TPV model effective for them. However, large enterprises conduct more spending through bill payments and procurement systems, causing the share of card spending to level off. Ramp's core capability is reducing bureaucracy, complexity, and wasted time — this delivers immense value to large enterprises, yet the TPV model fails to capture it.
"We drive a lot of value, but we can't capture it through card transactions. So we need to charge for software. Moreover, charging gives us better market feedback signals — which products customers are willing to pay for." (Karim Atiyeh)
The result of the transition: growth not only remained intact but accelerated — the sales team became more motivated to promote these products, customers became more engaged, and Ramp gained better visibility into its roadmap.
Thoughts on AI pricing: Atiyeh is cautious about token-based or time-based pricing models, believing they could incentivize engineering teams to avoid efficiency. "I'm still not sure what the final form will look like."
Karim Atiyeh believes that technical founders hold an advantage in the AI era because they can see possibilities that non-technical experts cannot.
Atiyeh uses the Ford automobile analogy: if you ask people what they want, they will say "a faster horse," not a car. Technical founders know what an engine can do, so they can envision the car. At the same time, AI is narrowing the expertise gap — "I am more like a doctor now than ever before, but I am not a doctor. I just know what questions to ask."
Practical case: CTO takes over marketing: When Ramp's conversion rate and monetization ability approached their upper limits, Atiyeh found that the leading indicator of marketing (lead generation) was slowing down. The core changes after he took over marketing included:
Classic case — dynamic billboards: A billboard in Union Square costs $100,000, but changing its content costs only $1,000. Changing it seven times a week brings the total cost to $107,000, with an effect far exceeding that of two static billboards ($200,000).
| Position | Guest View | Key Data |
|---|---|---|
| Amex | Risk warning (strong brand but weak product) | Early Ramp's pitch was "no worse than Amex" |
| Concur | Risk warning (poor user experience) | Used as a negative case study |
| Capital One | Positive (acquired Paribus) | Product renamed to Capital One Shopping post-acquisition |
| AWS | Positive (infrastructure) | Paribus runs on AWS; previously communicated with the AWS team |
| Visa/MasterCard | Neutral (strong business model but widely misunderstood) | Payment costs mainly come from consumer rewards, not Visa's cut |
| Stripe | Positive (customer) | One of 70,000+ customers |
| Cognition | Positive (case study) | Its launch video garnered significant attention |
| Dyson | Positive (philosophy) | Pursuing differentiation itself as a strategy |
1. "Your code is an LLM plus instructions and an infinite loop" (Karim Atiyeh) — The second phase of AI involves programming LLMs into the product itself, rather than merely using them as auxiliary tools. Ramp's policy agent is a practical embodiment of this philosophy.
2. "You can know exactly how long something will take, or you can do it very fast — it's hard to have both" (Karim Atiyeh) — He opposes detailed estimation before development, arguing that it incentivizes teams to provide longer estimates, thereby reducing speed.
3. "Celebrating lagging indicators jeopardizes the urgency to solve current problems" (Karim Atiyeh) — The results of the current quarter are determined by work done 6–12 months ago, so today's numbers should not be celebrated.
4. "Technical founders can see possibilities that others overlook" (Karim Atiyeh) — Using the Ford automobile analogy: people only ask for faster horses, but technical experts know what an engine can do. AI is narrowing the expertise gap, enabling technical founders to solve problems across non-technical domains.
5. "One billboard changed 7 times is more effective and cheaper than two static billboards" (Karim Atiyeh) — Systems thinking in marketing: the cost of modifying existing assets is far lower than acquiring new ones, so the iteration frequency of existing assets should be maximized.
6. "Look for 'spikes' rather than 'full-stack'" (Karim Atiyeh) — When hiring, only deeply examine the area the candidate claims to excel in, rather than checking boxes across 10 dimensions. This uncovers extreme talent rather than average talent.
7. "If no one is trying to kill you, you're probably not doing enough right" (Karim Atiyeh) — During the Paribus era, the company received legal threats from multiple retailers, which actually validated the product's value. The best time to invest may be right after a company has been attacked and survived.
8. "The reward for solving a challenge is just a more complex challenge" (Karim Atiyeh) — Therefore, enjoying the journey itself and working with great people is more important than pursuing the destination. Ramp's ultimate vision is to make users never need to log into Ramp at all — "autopilot finance."