Summary

Matt Biilmann’s agent-economy thesis is expansion, not developer extinction: agentic tools could create the next 100 million software developers. As Squarespace and similar tools lowered the barrier to website creation, and boot camps added roughly 14 million developers at the React layer without eliminating lower layers, tools such as Bolt, Lovable, and Cursor should add another professional layer. “There’ll still be a skill set involved” in deciding what to build and pushing agents to their limits.
“Agent experience” means redesigning products so agents can complete work before their human users perform traditional setup. Netlify’s flow lets an agent create and deploy a site without registration, return a working URL, and offer account claiming or a custom domain afterward. Yet Biilmann insists the endpoint remains human: “I don’t think anyone can have a great developer experience five years from now if they don’t have a great agent experience.”
Generic SaaS faces a credible five-year threat as AI collapses the cost of custom internal software, though the host disputes the breadth and speed of substitution. Biilmann moved from a 10–15-year timeline after seeing Bolt and Lovable advance, expecting budgets could shift from SaaS subscriptions into internal tooling or custom providers. The thesis only works atop zero-operations, serverless, managed components; otherwise maintenance “just overtakes everything.”
Salesforce could lose workflow ownership while retaining an exceptionally valuable position as the system of record beneath agent-built applications. As custom UIs proliferate, durable value may sit where data, permissions, authentication, rules, and auditing live. Biilmann calls reduction to a repository Salesforce’s “big threat,” but warns against underestimating a foundation on which new workflows run.
Chat is an entry point, not the destination, and AI raises rather than removes the premium on differentiated experience and taste. Figma could face an existential threat to its current revenue model if prototyping migrates directly into prompted, working software, but brands still need visual storytelling. Once anyone can clone today’s standardized storefront, winners must create experiences that “guide me and…take me somewhere”—something an empty chat box cannot do.
Biilmann favors open agent ecosystems because builders can swap and combine models, while the host sees formidable closed advantages at Google and Microsoft. Falling training costs, open models, and DeepSeek weakened the case for a few vertically integrated model owners; meanwhile Google’s consumer endpoints, chips, and compute, and Microsoft’s enterprise security remain powerful. Biilmann concedes walled gardens will persist but says their current integrations produce “more bland and worse” experiences.
AI spending should come from labor, development, operations, and maintenance budgets rather than requiring an entirely new wallet. Call centers already demonstrate the shift, with resolution pricing appearing instead of per-seat pricing; Stripe and other zero-operations building blocks could benefit as more custom products and internet businesses emerge. Given Anthropic at $60 billion, OpenAI at $300 billion, and Groq at $50 billion, Biilmann would buy Anthropic and sell Groq; for one public stock over 10 years, he chooses Microsoft.

Deep dive

1. Agent experience removes setup before it removes interfaces

Biilmann starts from two possible futures: agents either “escape” and leave humans behind, or remain tools acting because a person wants something accomplished. His discipline of agent experience assumes the latter and asks what happens when a user reaches a product through an agent rather than its normal interface.
Netlify’s concrete redesign lets a code-generation agent build and deploy a Sudoku app, receive a public URL, and show the result without its user first opening an account. Only afterward can the user claim the site, attach it to an existing account, keep it around, or add a custom domain.
The strategic instruction to product teams is empirical: identify which agents humans actually use with the product. For Netlify that includes Cursor, Copilot, and Windsurf at one layer, then Bolt, Lovable, Create XYZ, and Refine at another; other industries will have different agent surfaces.
Traditional UX does not disappear because humans still judge the holistic outcome. Biilmann’s sharper claim is that “no one can have a great developer experience five years from now if they don’t have a great agent experience,” even though interaction flows and information presentation will change radically.

2. Custom software shifts the build-versus-buy boundary

Biilmann’s SaaS argument rests on altered economics: today, hiring developers to build, deploy, and operate a bespoke internal app rarely beats buying a generic product. As agentic development collapses that cost, solving the company’s exact problem becomes viable across a growing share of workflows.
Generic SaaS encodes shared best practices, but it also forces customers to adapt their operations to somebody else’s product. Biilmann recalls repeatedly changing Netlify’s processes because “this tool kind of says we should do things this way”; cheaper custom development reverses that accommodation.
The host pushes back that claims such as Klarna replacing all SaaS with AI are not credible, and that slow-moving enterprises will not rebuild hundreds of systems soon. Biilmann says he shared something close to that view a year earlier, but Bolt and Lovable shortened his expected horizon from 10–15 years to roughly five.
Adoption should begin where no-code products such as Retool already sit, then spread through internal-tooling teams and outside custom builders. Biilmann expects some budget to move from SaaS subscriptions into those teams, while stopping short of saying every standardized provider disappears.

3. Systems of record and zero-operations infrastructure become the durable layer

Salesforce’s danger is becoming “some kind of system of record” instead of the workflow platform, but Biilmann is ambivalent about whether that is ultimately bad. If agent-built applications all depend on its data, business rules, authentication, permissions, and auditing, the repository itself remains highly valuable.
Maintenance is the constraint on replacing 200 SaaS tools with 200—or 400—specialized applications. A code agent cannot hand an enterprise “a massive Kubernetes cluster” and say, “Good luck”; fully custom operational stacks would destroy the economic case.
The viable architecture composes maintained “Legos”: zero-operations deployment, serverless APIs, managed content systems, managed databases such as Superbase or NeonDB, cloud services, and systems of record such as Salesforce. Depending on agents themselves to maintain bespoke infrastructure belongs, in Biilmann’s estimate, to a 15–20-year horizon.

4. AI makes current interfaces cheaper before inventing new ones

Figma would need to connect visual design to the full chain of prototyping and building because projects are already starting with prompts instead of templates, framework bootstraps, or design prototypes. As long as brand and visual craft matter, teams will still work at the visual and interface level; disconnection from executable output could become “a pretty existential threat” to Figma or its current revenue model.
Chat remains useful for the same reason Slack, Messenger, and WhatsApp are useful, yet Biilmann rejects it as the “be-all and end-all.” Most AI builders currently produce the same websites, dashboards, and marketing experiences made over the previous decade; they have lowered production cost without discovering the new interface.
The next frontier exploits what deterministic computers could not: generated 3D assets, sound, music, and visuals inside interactive experiences. Generative AI removed the underlying constraint that every click must yield a predictable, predefined outcome, but product designers have barely explored the resulting design space.
The host asks why ecommerce cannot simply resemble entering a store and requesting an item. Biilmann points to what a fancy brand spends on display windows, environments, and people who draw customers in before they ask; an online brand likewise needs to “guide me” rather than present a box saying, “You come up with what you want me to do.”

5. The next 100 million developers still form a profession

The host invokes Webflow and Squarespace’s promise that anyone can build, then says his mother still cannot use Shopify, Webflow, or Squarespace; he says she can use Lovable. Biilmann agrees that the tools lower barriers but rejects the leap from easier tooling to universal self-service: restaurant owners may still prefer paying a specialist because they have a business to run.
His best evidence is a Niantic UX designer who prompted Bolt into building a browser-based, AI-assisted 3D object editor. The result is “mind-blowing,” but Biilmann doubts either he or the host’s mother could reproduce it—the tool still requires accumulated skill in deciding what to build and how to push it.
Earlier site builders did not end web development; they let every corner shop have a site while raising the standard for professional work. Later, boot camps created roughly 14 million developers at the React layer atop Ruby on Rails, Java, and .NET layers whose populations remained at least as large.
Biilmann predicts the same layering again: perhaps 100 million new developers operating mainly through agents, with some descending further into the stack. Developers who fully understand the lower layers will remain necessary and will likely think about the agent experience of what they build; agent-native specialists will learn “what the agent should actually build” and how to reach it fastest.

6. Open agents compete by routing around closed model stacks

Biilmann initially saw large language model makers as candidates for vertically locked platforms, but cheaper training, open models, and DeepSeek changed the structure. Agent builders now benefit from switching among whichever models perform best and combining models instead of accepting one provider’s full stack.
The host challenges that conclusion with Google’s consumer distribution, chips, compute, and G Suite endpoints, plus Microsoft’s enterprise reach and non-trivial security advantages. Biilmann concedes that open victory does not eliminate closed gardens, just as the open web coexists with them.
His present signal nevertheless favors openness: Genesis buttons appear throughout Google products, yet he never uses them, while a Chrome extension can bring ChatGPT—with its history and familiarity—into Google Docs. Vertical integrations currently feel “more bland and worse” than the competing agent experiences built outside those suites.
Model specialization is already beginning. Biilmann highlights Anthropic’s programming strength because code lets a model make other computers perform tasks, multiplying its reach; mechanisms originally developed for safe outputs also help produce syntactically valid, production-usable results.

7. Value migrates from labor and operations into agents and primitives

Bolt, Lovable, Cursor, and Windsurf can absorb money previously spent on developer hours, while platforms such as Netlify can absorb maintenance, DevOps, and operating costs. The economic premise is not that every layer claims fresh budget, but that customers build more things with expenditure redirected from old production inputs.
Call centers provide the clearest labor substitution: companies are willing to move staffed hours onto LLMs, and the host notes that the commercial model is shifting from per-seat charges to resolution pricing. Biilmann expects some labor to shift into creating agentic workflows, applications, and products.
The host argues Stripe could become far more valuable as newly enabled entrepreneurs create stores and expand “the GDP of the internet.” Biilmann agrees because Stripe is an abstract, zero-operations transaction primitive that agents can compose into custom systems, unlike a ready-made end-to-end ecommerce shop.
Asked to allocate among Anthropic at $60 billion, OpenAI at $300 billion, and Groq at $50 billion, Biilmann buys Anthropic for production-usable coding and sells Groq; OpenAI retains the consumer reflex and the ingrained association of ChatGPT with talking to an AI. His 10-year public holding is Microsoft, chiefly because it should benefit from disruption and he regards Satya as the strongest CEO in that space.

8. Founder hindsight tempers the closing AI optimism

Biilmann calls WordPress late in its innovation cycle and poorly positioned for the disruption, expecting it to look increasingly like legacy software. It might ossify through community persistence or face some kind of PE exit, but he struggles to see venture-scale acceleration returning.
Netlify’s hardest financing was its seed: around 100 meetings, or “100 or something,” involving angels, individuals, and funds, produced $2.135 million at an $8 million post-money valuation. Its last round, at the end of 2021, valued the company at $2 billion; Biilmann says that market was “priced too high” and not entirely healthy, while still believing founders should capitalize companies when favorable terms provide resilience.
Asked for a public-market short, Biilmann says he is bad at betting against things but offers ExxonMobil as a possibility because “if they go down, it’s probably a good thing.”
His most surprising AI-era behavior was the initial safety panic over LLMs immediately seizing nuclear weapons or outcompeting humanity. He saw systems still far from independent agency and considered strict early regulation premature; as the host jokes, “It doesn’t even do a Twitter thread.”
Against predictions that agents eliminate websites and applications, Biilmann believes humans will continue needing shared, linkable experiences where “I’m seeing the same as you’re seeing.” His hoped-for outcome is open tooling that lets one illustrator make AI-generated TikTok sci-fi epics or one designer build a 3D editor—and lets creators own what they make.

Apple's Siri-ous Problem + How Starlink Took Over the World + Is AI Making Us Dumb?26. Behind the Manus Controversy: FOMO, Ambition, and Loss of Control—but the Age of Everyone as a Leader Is Coming!