The Twenty Minute VCKlarna CEO: SaaS is Dead: Why Systems of Record Will Die in an Agentic World
CHAPTERS
AI makes software creation nearly free—and that’s the start of the SaaS threat
Sebastian argues the cost of building software is collapsing toward zero, which changes where durable value can exist in software businesses. He frames today’s disruption as only the first wave: code generation is cheap, and the next shock will be moving and reshaping data across systems.
- •Software creation costs are rapidly approaching zero due to AI coding tools
- •Competitive advantage shifts away from “we can build it” toward other moats
- •Current disruption is mostly about code generation—not yet about data portability
- •The bigger coming change is reducing switching costs tied to locked-in data
The real SaaS danger: one-click data migration and collapsing switching costs
The core moat for many SaaS systems is that customer data is embedded in proprietary models and workflows. Sebastian predicts AI agents will soon automate extraction, mapping, and migration—making it far easier to swap vendors and pressuring incumbents’ valuations.
- •SaaS lock-in relies heavily on data models and embedded workflows
- •Agents will automate data extraction/mapping to new tools
- •Lower switching costs increase churn risk even if products remain “good enough”
- •The stock market is beginning to price this risk into SaaS leaders
Where software multiples go: from high-growth SaaS to utility-like pricing
Sebastian discusses how public-market multiples may compress further as software becomes more substitutable. He contrasts historical SaaS price-to-sales ranges with utilities and cites extreme examples like Chegg to illustrate how AI shocks can re-rate categories.
- •SaaS traded at ~20–30x sales historically; many now at ~5–10x
- •Utility-like businesses often trade ~1–2x sales, suggesting downside risk
- •Chegg is used as a cautionary case of AI-driven value destruction
- •He expects re-rating pressure but not necessarily “overnight” collapse
From “build everything” to “Company-in-a-Box”: open-source + agents as the new stack
Harry challenges whether enterprises will really rebuild internal tools. Sebastian responds that the future may be standardized Lego-like components (often open source) stitched together by agents, producing broad, integrated “company-in-a-box” experiences for smaller firms and new operating systems for larger ones.
- •AI will increasingly assemble reusable components rather than generate everything from scratch
- •Open-source modules + an agent layer can replicate many back-office workflows
- •Small businesses will buy integrated bundles rather than “vibe code” internally
- •Winners may be broader, more integrated suites rather than siloed point solutions
Why Klarna pulled back from SaaS: context unification for an AI-native operating system
Klarna’s rationale for reducing SaaS usage is to give AI the best possible context. Sebastian argues siloed tools make it harder for agents to understand the full business state, so Klarna is reimagining its stack as an AI-first ‘operating system’ blending deterministic and probabilistic software.
- •AI performance depends on unified, high-quality context across the business
- •SaaS silos fragment data, policies, and workflows across many systems
- •Klarna is rebuilding toward an AI-native bank operating system
- •Some SaaS remains (e.g., Slack), but core context is being consolidated
Why large companies may need bespoke AI customer service (and why Klarna built it)
Customer support looks crowded with vendors, but Sebastian says truly effective AI support needs deep, accurate context—often the source code itself. For Klarna, the support agent becomes part of the tech stack, which is why off-the-shelf tools weren’t sufficient.
- •Klarna’s early AI support handled simple queries and scaled quickly
- •High-quality answers require ground-truth context (often in source code)
- •Documentation can be wrong; code is the ultimate reference
- •At scale, customer support AI becomes a core internal platform capability
The PR backlash and the “VIP future”: humans as premium service + Klarna’s Uber-like support model
They unpack how headlines about replacing agents created backlash and confusion. Sebastian argues AI will commoditize basic support, while human interaction becomes premium; Klarna is testing a marketplace model recruiting passionate customers for part-time support to improve satisfaction.
- •Media simplified the story into “Klarna replaces humans,” causing backlash
- •AI will become the cheap default; human service becomes VIP/premium
- •Klarna recruits its own customers for part-time support (an “Uber model”)
- •Higher satisfaction comes from product-native, motivated human helpers
Labor displacement, CEO honesty, and the coming organizational shrink
Harry raises investing around job replacement; Sebastian agrees displacement is real and says many leaders avoid admitting it. He explains Klarna’s headcount drop (roughly halved) via attrition and predicts further decline, while emphasizing relationship roles will remain.
- •Sebastian aligns with candid “displacement will happen” messaging
- •Klarna reduced headcount dramatically, largely via attrition and hiring restraint
- •He predicts potentially <2,000 employees by 2030
- •Relationship-heavy roles (merchant partnerships, human connection) persist
AI changes boardroom math: shipping more products without asking for more budget
Sebastian describes how AI altered investment constraints: Klarna could expand into new banking services while shrinking costs, making board approval easier. He also discusses employee incentives—sharing productivity gains via higher compensation per head.
- •AI enables new product launches without proportional headcount growth
- •Boards are more receptive when expansion doesn’t require major new spend
- •Klarna tied efficiency gains to higher employee compensation per head
- •He frames this as creating safety and buy-in during transformation
Fintech endgame: digital financial assistant, US scale, and competing with Revolut/Nubank
Sebastian revisits a 2015 strategic vision: banking becomes a proactive digital assistant. He argues Klarna’s edge is scale, brand, and uniquely rich purchase-level data from its payment rails—critical for advice and personalization—while emphasizing the US as necessary for global relevance.
- •Vision: a proactive assistant that saves users time and money in daily finances
- •US is strategic for global scale; otherwise you risk being acquired by US players
- •Klarna’s rails capture item-level receipts (richer than card transaction data)
- •He positions Klarna as lifestyle/Amex-like vs Revolut’s different entry point
Valuation lessons from 2021: when multiples outrun fundamentals and hiring gets painful
They discuss the dangers of high valuations and rapid multiple expansion. Sebastian’s key reflection is that over-hiring during euphoric markets makes later layoffs more likely, and leaders should watch when valuation multiples expand faster than revenue growth.
- •Risk signal: valuation multiples expanding faster than revenue growth
- •He regrets not being more cautious on hiring ahead of downturn
- •Layoffs after aggressive hiring are emotionally and culturally costly
- •High valuations can distort decisions and expectations across stakeholders
How Sequoia invested, Moritz joined the board, and what great investors do differently
Sebastian tells the origin story of courting Sequoia from Stockholm and how a bold comment led to Michael Moritz joining Klarna’s board. He credits Moritz’s ability to distill massive information quickly and notes specific inflection calls (e.g., ‘now or never’ on US expansion).
- •Sequoia’s early investment was a major credibility boost in talent markets
- •A deliberate ‘cocky’ moment helped secure Moritz’s board involvement
- •Moritz excels at identifying what matters from complex information
- •Board-level conviction helped trigger sharper US focus at a key moment
Investors who don’t build will lose: why VCs must use AI tools themselves
Sebastian argues many investors are funding AI without understanding capabilities or differentiation. His advice is simple: personally build with tools like Cursor/Claude Code to judge what’s real, what’s commoditized, and where moats can exist.
- •AI investing requires hands-on familiarity with current tooling
- •Non-building investors misjudge differentiation, cost curves, and timelines
- •Tool performance varies by task; ‘personalities’ and workflows differ
- •He expects ongoing commoditization pressure across developer tools
What CEOs really think about AI: adoption is slower than capability, enterprise lags consumer
In the quickfire and closing themes, Sebastian says he’s updated his view of the adoption curve: behavior change takes time, especially in enterprise. He also shares how public scrutiny changes CEO time allocation and how narrative whiplash follows valuation cycles.
- •Biggest update: adoption pace is slower than the tech’s raw capability
- •Consumers adopt faster; enterprises change habits and processes more slowly
- •Public-company life increases time spent on investor communication
- •Public perception flips from ‘genius’ to ‘disaster’ with valuation cycles
AI as compression: fewer duplicated systems, and an uncertain future for compute demand
Sebastian frames AI as a compression technology that collapses duplicated knowledge and messy enterprise data into fewer sources of truth—like Wikipedia’s discipline at scale. He debates whether enterprise compression reduces compute needs more than entertainment-generation increases them, referencing a discussion with Michael Burry.
- •AI compresses repeated information rather than storing duplicates
- •Enterprise data is highly redundant across tools (docs/CRM/chat/code)
- •Compression could reduce future compute needs for many enterprise workloads
- •Counterforce: personalized media/entertainment generation could drive compute up
