Claude Code Ends SaaS, the Gemini + Siri Partnership, and Math Finally Solves AI | #224

The panel discusses Vlad Tenev's concept of a 'job singularity' and the emergence of micro-corporations. Peter has predicted his children wouldn't need driver's licenses or university degrees; he runs workshops to help teens 'go from future shock to future shape'. Salma emphasizes that the psychological transition from consumer to creator is vital. Dave argues college may still serve for networking but not for job credentials.

Peter describes how multiple Opus 4.5 agents create code so rapidly that he can't keep up mentally, unlike the old days when slow coding allowed continuous architectural thought. He jokes that he now revisits a GUI and asks, 'What did I call this again?'. Dave mentions ex-Yahoo developers are stunned. The panel calls this a transition phase of the singularity where cognitive demands peak, even as productivity soars. Survey research confirms AI masters feel extreme weekly stress but produce unprecedented output.

The panel discusses a slide showing tombstones for SaaS and no-code companies, joking that Claude Code can rebuild Salesforce or Stripe from a prompt. Salma argues a new AI-native enterprise stack will emerge alongside the legacy one, rewriting business logic from scratch. Dave notes 'Only the paranoid survive' and that quant funds now track talent movement as a predictor. Alex adds that CRMs are already heavily customized and low-compliance, so AI-generated alternatives will eat away gradually, but the market will find equilibrium.

Alex argues that the time to superintelligence is so short that legacy environmental or safety concerns should be deprioritized. Dave adds that even if AI discovers fusion soon, turbines and generators are sold out, so near-term energy abundance must come from solar, nuclear, and fossil. Peter notes that 20 African countries now import gigawatts of solar panels from China, showing the geopolitical dimension of energy. The panel bemoans that regulatory systems remain stalled by voter fear, despite rational arguments.

Alex explains that the combination of Opus 4.5 and Claude Code is achieving 'absurd amounts of autonomy', such as generating entire web browsers with JavaScript engines from scratch. This follows a hyper-exponential trend in autonomy time horizons, where tasks that once took years can now be done in days. Peter notes that writing code is paradoxically more taxing because you launch many agents and have to track dozens of outputs at once; he's generating more code in months than in his entire prior career. The mental load and cost ($100–$1,000 per day) are high, but productivity has skyrocketed. Dave's former Yahoo co-workers are 'walking around with their jaws dropped'. The panel debates whether established SaaS companies will be crushed or can survive by having access to the same tools. Salma argues a new AI-native enterprise stack will emerge alongside the legacy one.

Alex argues that while Cosmos is a big deal, it won't completely eliminate the value of real data: long-tail rare events still need real-world capture, especially in compliance-heavy domains. Dave counters that Nvidia's synthetic world models are helpful but still can't model niche physical scenarios like fusion reactor containment or nano-scale surgery; plenty of specialized spatial AI data remains. Peter sees Cosmos plus Vera Rubin as Nvidia becoming 'the AWS of reality'. They also discuss the DRAM shortage: memory demand is infinite, not cyclical, and Elon Musk is building his own fabs in response. This sets the stage for a decade of compute scarcity and high GPU prices.

Scott Stanford's framing: 'We move from a search box that gives information to a magic box that gives action.' Alex adds contrarian nuance: UCP is a JavaScript-based standard for e-commerce within chat agents, not the extinction of the web. Many purchases remain browse-oriented rather than conversational. Moreover, habit inertia is powerful; moving a send button a few pixels once tanked Yahoo Mail usage. So while UCP is important, the web won't vanish overnight. The panel agrees that universal typing might decline, and a verbal + AR interface may dominate, but reading and browsing will persist.

Alex explains that this moment will be remembered as the inflection point when everything started to get solved by AI. Math was the easiest starting point because problems are straightforward to enumerate and verify. The same approach will walk out into other sciences. Peter adds that the constraint is not problem difficulty but access to data, guardrails, and evals. Whoever first brings the necessary data and regulatory framework to a new field can unlock massive value — the next Meror (referencing the platform that set the standard for AI evaluation). Salma is struck by 'problems wait to be prompted', noting that it means our only limitation is what we can imagine to prompt the AI to solve.

Salma predicts consulting firms will do well because in a volatile world, clients need help, and being half a step ahead is enough. However, he scoffs at the 1:1 ratio, arguing that systems with exo-agents crawling through companies can easily reach 100 agents per human. Alex raises an ironic possibility: if agents are counted as 'per capita', productivity statistics may appear artificially suppressed. Peter notes a risk: consulting firms' clients themselves may not survive the seismic shock, making this the biggest advisory opportunity in history as all institutions need rebuilding. The panel concludes that the old business model must change, likely toward shared-value outcome pricing.

Peter and Salma have been discussing this for months and now see it confirmed. They emphasize that by the time today's teenagers finish school, the concept of employment will be unrecognizable. Dave, who has a 14-year-old son, predicts college won't be necessary for a job, only perhaps for networking or starting a company. Peter runs workshops with teenagers to prepare them for this reality. The panel unanimously agrees that 'the future job is entrepreneur' and that clinging to employee mindsets is dangerous.

Alex points out that when GPT-5 launched with reasoning on by default, many users disliked the slower, less sycophantic personality and chose not to use reasoning features. This means forcing expensive inference is not guaranteed. Enterprises use reasoning, but not enough to triple revenue year over year. The 'elephant' is that the capex needs transformative, sticky applications that haven't yet materialized at scale. Dave notes the shift to heavy infrastructure feels more sustainable, but the panel agrees that we are in a field-of-dreams moment: build it, but will the revenue come? The outcome will determine which frontier labs survive.