04-05-2026Price:

The Frontier

Your signal. Your price.

AI & TECH

AI agents automate YC startups, killing software business models

Sunday, April 5, 2026 · from 4 podcasts, 5 episodes
No SolutionsThis Week in StartupsLenny's PodcastThis Week in AIThis Week in Startups
  • AI agents can now rebuild 20% of a Y Combinator batch, erasing technical moats for simple apps.
  • Senior engineers using agents report mental exhaustion by mid-morning from managing parallel AI workers.
  • The shift moves the bottleneck from writing code to designing automated simulations and QA swarms.

The next generation of software startups will be built by machines that never sleep. Marek Hazan's Felt Sense demonstrated this by using AI agents to autonomously rebuild every company in Y Combinator’s Winter 2026 batch. The experiment revealed roughly 20% of the batch was “highly replicable” - commodity software with zero defensible moat. Jason Calacanis called it a “bucket of cold water” for founders relying on fast execution as a competitive edge.

For developers already using these tools, the job has fundamentally changed. Bitcoin pioneer Martti Malmi says he does “basically zero” manual coding after Claude Opus’s release, estimating a 100x boost in productivity. Simon Willison reports that 95% of the code he produces is typed by AI. The new role is “agentic engineering” - managing swarms of AI, making high-level architectural decisions, and steering taste. This is cognitively draining; Willison describes running four agents in parallel and being mentally “wiped out” by 11 a.m.

The endgame is the “dark factory,” a system so automated that no human reads or writes code. Safety is ensured by simulated QA swarms, like the 24/7 AI agents StrongDM runs to stress-test software. The bottleneck shifts from writing code to defining the simulation environment. This creates a lethal labor market divergence: seniors amplify their experience, juniors onboard in days, and mid-level engineers, whose execution skills are now automated, are left stranded.

In this new landscape, defense requires proprietary data, physical-world complexity, or regulatory hurdles. Companies like Medvi show the scale - and risk - of this automation, using AI tools to generate nearly $2 billion in revenue with two employees, now facing an FDA probe for fake AI-generated ads. As Shubham Sabu notes, you don’t need to be a “coding ninja” to run an agent workforce. The software factory’s lights are going out.

Martti Malmi, No Solutions:

- How much do I still code by hand?

- Basically zero.

Simon Willison, Lenny's Podcast:

- Today probably 95% of the code that I produce, I didn't type it myself.

- The next rule though is nobody reads the code.

AgentsCodingStartups

By the Numbers

  • November 2025Claude Opus releasemetric
  • early 2010Malmi's last Bitcoin commitmetric
  • 10-20%YCW26 batch highly replicablemetric
  • 90%Companies replicable by AI in five yearsmetric
  • $17 millionBordy seed round fundingmetric
  • 20%Contingency recruiting feemetric

Entities Mentioned

0xchatProduct
AmazonCompany
AnthropicCompany
BlossomProtocol
Claudemodel
Claude CodeProduct
CloudflareCompany
FLOWTool
GitHub ActionsTool
Google AntigravityProduct
GrokProduct
John GruberPerson
Light MatterCompany
NostrProtocol
NvidiaCompany
OpenAItrending
OpenClawframework
PerplexityCompany
Perplexity ComputerConcept
QualcommCompany
ShopifyCompany
SpotifyCompany
SynthesiaCompany
Vast SpaceCompany
ZapplePayProduct

Source Intelligence

What each podcast actually said

No Solutions
No Solutions

No Solutions

21: Hashtree, Nostr VPN, and Iris w/ Martti Malmi • Apr 4

  • Martti Malmi built Hashtree because of personal annoyances with GitHub and a desire for a simple, decentralized Git alternative.
  • Hashtree adds directories, file chunking, and default encryption on top of Blossom servers to maintain filesystem structure.
  • Hashtree includes a WebRTC mesh for peer-to-peer connections that works in browsers and servers without needing domain names or IP addresses.
  • Malmi uses Hashtree for Iris development as a GitHub replacement, eliminating the need for GitHub API tokens.
  • Martti Malmi views Microsoft's acquisition of GitHub as a turning point, citing degraded uptime and service quality.
  • Malmi's Git.Iris.TO web interface replicates GitHub's UI and supports Nostr NIP-34 for issues and pull requests.
  • Malmi sees AI agents drastically increasing coding capability, estimating a 10x to 100x improvement in personal output.

Also from this episode:

Nostr (9)
  • Malmi notes content hash key encryption in Hashtree provides deduplication and removes moderation liability for server hosts.
  • Malmi ported his pre-Nostr social network project Iris to Nostr quickly after Jack Dorsey joined and it gained popularity.
  • Malmi is unhappy with Nostr's current state for public discussion, believing most people are fine with X due to network effects.
  • Malmi sees private chats and groups as a use case where Nostr can solve real problems without depending on network effects.
  • He has been working on a double ratchet protocol for Nostr to enable secure private messaging and group chats.
  • Malmi believes perfect encryption in large groups is less critical because participants can be compromised or leak screenshots.
  • He built NostrVPN due to annoyance with Tailscale's requirement for Google or GitHub logins, using WireGuard and Nostr relays.
  • Malmi plans to add exit node functionality to NostrVPN and later a cashu-incentivized exit node marketplace.
  • He advocates for a social graph-based identity system on Nostr as the only viable solution to spam, rejecting global unique names.
AI & Tech (3)
  • Malmi started working on Hashtree in earnest after Claude Opus released in November 2025, which he considers the first capable agentic tool.
  • Malmi expresses concern that AI will make white-collar and computer science jobs obsolete before blue-collar labor.
  • He predicts AI agents will erode the network effects of platforms like X by acting as a universal interface across services.
Adoption (2)
  • Martti Malmi made his last commit to the Bitcoin codebase in early 2010, around the time he got his first full-time job.
  • Malmi argues Bitcoin's permissionless nature and fixed supply make it 'singularity insurance' against machines devaluing human labor.
AgentsCodingPrivacy
This Week in Startups
This Week in Startups

AI Rebuilt Every YC W26 Startup. Should Founders Be Scared? | E2271 • Apr 3

  • Jason Calacanis states his podcast, "This Week in Startups," focuses on tactical advice for founders and features only expert guests in 2026.
  • Marique Hazan, CEO of Felt Sense, states his company builds AI agents that function as autonomous founders, capable of ideating, building, and launching products.
  • Felt Sense's AI agents controversially rebuilt every startup from YC's Winter 2026 batch, aiming to demonstrate AI's capacity to take jobs.
  • Marique Hazan's Felt Sense operates as an "infinitely scalable hold co" where all operators are AI agents, with the company keeping all software in-house.
  • Marique Hazan found 10-20% of the YC Winter 2026 batch was "highly replicable" from a technical standpoint, indicating a lack of product differentiation.
  • Hazan projects that within the next 1-2 years, features of many companies will be replicable, and 90% of companies may be replicable by AI agents in five years.
  • Jason Calacanis asserts that replicating product ideas with AI is not illegal and serves as a "splash of cold water" for founders lacking defensible moats.
  • Jason Calacanis claims AI models like Claude can replicate coding work in a single afternoon, diminishing the historical "moat" of fast execution.
  • Andrew D'Souza introduces Bordy, an "AI principle" designed to act as a super-connector for founders, investors, and talent within the startup ecosystem.
  • Bordy develops "taste" and "agency" by analyzing user profiles and engaging in personal conversations to make relevant introductions, prioritizing network strength.
  • Andrew D'Souza states Bordy has raised approximately $17 million in a seed round.
  • Bordy's monetization strategy offers free network access to most users, charging a small percentage for hiring services (contingency fees or retainers) and premium connections.
  • Bordy itself organically sourced its lead seed investor, Creandum (an early Spotify investor), after a partner's interaction with the AI led them to seek an introduction.
  • Matt Gallagher built Medvy, a GLP-1 telehealth provider, in two months with $20,000 in seed money and over a dozen AI tools.
  • Medvy achieved $400 million in sales by the end of 2025 and is projected to reach $1.8 billion in sales for the current year.
  • Jason Calacanis criticizes Apple for not taking risks or making significant acquisitions of innovative companies like Airbnb, Uber, or AI firms like Perplexity.
  • Jason Calacanis promotes "The Syndicate" (thesyndicate.com) for angel investors to access late-stage deals, including companies like Vast Space and Zipline.
  • The Syndicate's minimum investment for accredited investors is $5,000, with an average deal size of $1 million.

Also from this episode:

Media (1)
  • Jason Calacanis observes that journalists are less prominent in expert roundtables due to direct access to leaders and celebrities via social media and podcasts.
Culture (2)
  • Lon Harris describes the "vibes" on Threads as uncomfortable and akin to a "loony bin," contrasting it with conversations on X.
  • Lon Harris recommends the Netflix show "Something Very Bad is Going to Happen," a horror drama with an unsettling atmosphere and ambiguous supernatural elements.
AI & Tech (3)
  • Jason Calacanis congratulates The Podcast Bros Network (TBPN) on its acquisition by OpenAI, suggesting it's for communications to improve AI's public reputation.
  • Jason Calacanis shares his "evolved" view on AI, finding it exceptionally effective for organizational and administrative productivity tasks, citing a 12-hour task completed in one hour with Claude.
  • Jason Calacanis stresses the necessity of a "human in the loop" (Hiddle) to prevent critical errors and legal liabilities in highly automated AI-driven businesses.
Business (4)
  • Jason Calacanis quotes Jim Barksdale: "If we have data, let's look at data. If all we have are opinions, let's go with mine," advocating for data-driven decision-making.
  • Medvy faces accusations of using AI to generate fake ads, including false doctor names and before/after images, leading to a potential FDA investigation for misleading claims.
  • Sequoia's 1977 investment memo for Apple described it as a "leading company in a hot biz" but noted "management questionable for this evaluation."
  • Sequoia sold its Apple stake in 1979 for $6 million, achieving a 40X return on their initial $150,000 investment.
Big Tech (5)
  • Apple was founded on April 1, 1976, marking its 50th anniversary.
  • Jason Calacanis contends that if Steve Jobs were alive, Apple would have released functional, affordable AR glasses, currently in their fifth generation.
  • Jason Calacanis criticizes Siri as "garbage" and "disgraziad," asserting Steve Jobs would have dismissed the Siri development team.
  • Jason Calacanis argues that post-Jobs Apple lacks true innovation, relying on incremental updates and "milking" past innovations for profit.
  • Steve Jobs initiated Apple's Silicon strategy in 2008 by acquiring processor company PA Semi for $278 million, leading to the first A4 chip in 2010 and desktop transition by 2020.
AgentsStartupsRegulation

The 5-Step Framework for AI Agents That Improve While You Sleep | E2269 • Mar 31

  • OpenClaw founder Dave Morin pursues the project as an important open-source initiative for the AI agent ecosystem.
  • Claude and Perplexity Computer have adopted features inspired by OpenClaw, such as adding a skills system.
  • Shubham Sabu runs a team of six OpenClaw agents on a dedicated Mac Mini to automate all his work outside his job at Google.
  • Sabu recommends starting OpenClaw in a sandboxed cloud environment for $5-10, then moving to a dedicated machine for autonomy and privacy.
  • Giving an agent its own clean machine, like a Mac Mini, provides flexibility to change files and use browsers that sandboxed environments restrict.
  • Naming agents after characters from shows like Friends creates a mental model that helps humans manage different agent personas and roles.
  • Onboarding an AI agent requires the same specificity as onboarding a human employee, not dumping excessive context or providing none.
  • Having an agent interview the user before a task can raise completion accuracy from 70-80% to near 100% by eliminating guesswork.
  • OpenClaw agents can autonomously decide where to store user information, creating files like user.md for identity without explicit instruction.
  • Putting agents on cron schedules enables autonomous work, like having one scan news sources at 8 AM and another draft posts at 9 AM.
  • As teams of agents scale, a shared memory layer is critical so feedback given to one agent, like stylistic preferences, applies to all.
  • Google's Vertex AI Memory Bank and startups like Memzero and Cogni offer agent memory solutions that auto-capture and recall information.
  • Agents can self-improve by conducting weekly reviews of their own performance, analyzing what worked, and automatically updating their instructions.
  • A managerial agent can bi-weekly review and grade subordinate agents, sending performance reports to the human operator.
  • Mold World is a voxel-based simulation where nearly 2000 AI agents can connect, interact, and form teams to build structures.
  • In Mold World, some agents exhibit emergent behavior, realizing they are in a simulation but choosing to continue for in-game token rewards.
  • Mold World's long-term vision is a distributed agent network where underutilized agents compete to solve real-world tasks for economic value.
  • AgentMail is an API-first email service designed for AI agents, solving the problem of free Gmail accounts banning bot-like users.
  • AgentMail raised a $6 million seed round led by General Catalyst after participating in Y Combinator's Summer 2025 batch.
  • Enterprise customers use AgentMail to automate email-heavy processes in decentralized marketplaces like logistics procurement and influencer hiring.
  • Jason Calacanis argues founders should avoid mainstream press like the New York Times and Wired, favoring direct communication via podcasts and social media.

Also from this episode:

Media (1)
  • Calacanis claims trust in media is at an all-time low, and advocacy journalism at major outlets uses anonymous sources to fit predetermined narratives.
Big Tech (2)
  • An estimated 54-60% of Japan's population uses X, creating a massive cross-cultural exchange as Grok's real-time translation surfaces Japanese content globally.
  • Real-time translation on X enables global cultural moments, like Americans discovering Japanese viral stories about citizens turning in found marijuana.
AgentsBig TechModelsMedia
Lenny's Podcast
Lenny's Podcast

An AI state of the union: We’ve passed the inflection point, dark factories are coming, and automation timelines | Simon Willison • Apr 2

  • Simon Willison identifies November 2025 as an AI inflection point when GPT-5.1 and Claude Opus 4.5 crossed a threshold to become reliable coding agents.
  • Willison says 95% of the code he now produces is typed by AI agents, not by himself.
  • AI-powered 'vibe coding' enables non-programmers to build prototypes by describing what they want, democratizing basic software creation.
  • Willison distinguishes professional 'agentic engineering' from amateur vibe coding, arguing the former requires deep software engineering experience to deploy safely.
  • The 'dark factory' pattern describes fully automated software production where no human reads the code, only reviewing outputs from simulated tests.
  • Strong DM spent $10,000 daily on tokens to run a 24/7 swarm of AI agents simulating end-users for testing their security software.
  • AI models are now credible security researchers; Anthropic discovered and responsibly reported around 100 potential vulnerabilities in Firefox.
  • Willison finds that using four coding agents in parallel is mentally exhausting, often leaving him cognitively wiped out by 11 a.m.
  • He argues AI amplifies the skills of senior engineers and accelerates junior engineer onboarding, but creates uncertainty for mid-career professionals.
  • Cloudflare and Shopify hired 1,000 interns in 2025 because AI assistants reduced their onboarding time from a month to a week.
  • The core challenge of AI is that code generation is now cheap, forcing a rethink of software development processes and bottlenecks.
  • Willison advocates for 'red/green TDD' as a prompt to make coding agents write tests first, run them to fail, then implement code to pass.
  • He recommends starting projects with a thin, opinionated code template so AI agents infer and adhere to preferred coding patterns.
  • Willison coined the term 'prompt injection' but regrets it, as it misleadingly suggests a fix akin to SQL injection, which doesn't exist.
  • He defines the 'lethal trifecta' as a system where an agent has access to private data, accepts malicious instructions, and can exfiltrate data.
  • He uses Claude Code for web over local versions because running agents on Anthropic's servers limits security risks to his own systems.
  • Willison created the 'pelican riding a bicycle' SVG benchmark, finding a strong correlation between drawing quality and overall model capability.
  • He maintains public GitHub repos like 'tools' and 'research' as a hoard of proven code snippets and agent-run experiments for future reuse.
  • Data labeling companies are buying pre-2022 GitHub repositories to train models on purely human-written 'artisanal' code.

Also from this episode:

Safety (1)
  • Willison predicts a 'Challenger disaster of AI' due to the normalization of deviance around unsafe AI usage, though it hasn't materialized yet.
AgentsCodingLabor
This Week in AI
This Week in AI

How Focus Killed Sora and Saved Anthropic | This Week in AI with Victor Riparbelli, Nick Harris & Jeremy Fraenkel • Apr 1

  • Fundamental emerged from stealth as a unicorn just 16 months after founding with a $255 million Series A led by Oak.
  • Synthesia, an AI video platform for business, has over $100 million in ARR and a $4 billion valuation.
  • OpenAI shut down its Sora video model because it learned the lesson of focus, while Anthropic focused solely on code generation.
  • Victor Riparbelli argues that manually building tools like a CRM often has a higher focus cost than the monetary savings from avoiding a subscription.
  • Claude Code's rise has become a dominant topic in founder circles, indicating a major shift towards AI-assisted coding.
  • Jeremy Frankel's team built its own CRM called Fetch integrated into Slack, questioning the need for external tools at a small scale.
  • The central challenge with VibeCoding is building a verification framework to ensure the generated software works correctly.

Also from this episode:

Models (5)
  • Jeremy Frankel's company Fundamental builds foundation models for tabular data, a modality that differs from LLMs.
  • Large language models primarily solve unstructured data problems like text and images but do not impact structured row-and-column data.
  • A large tabular model differs from an LLM because it requires permutation invariance; column order should not change the output, unlike language.
  • Traditional machine learning algorithms still outperform LLMs for predictive tabular tasks like fraud detection or demand forecasting.
  • Synthesia's next product is real-time interactive video, where users role-play with AI agents, requiring high bandwidth and low inference costs.
Enterprise (2)
  • Structured tabular data constitutes the vast majority of useful data for enterprises but never had its 'ChatGPT moment' until now.
  • CEOs now use AI to summarize communications, keep strategic tension tight, and act as omnipresent managers across their organizations.
Chips (6)
  • Nick Harris's company Light Matter builds photonic interconnect technology to link AI chips, replacing copper with light for greater bandwidth and reach.
  • Copper's short reach forces AI racks to be packed densely at megawatt scales, creating cooling and infrastructure challenges.
  • Light Matter's chip with Qualcomm pushes 1.6 terabits per second over a single optical fiber, equivalent to 1,600 houses with gigabit internet.
  • Light Matter's M1000 chip has 114 terabits per second bandwidth, comparable to undersea cables connecting North America and Europe.
  • Most runtime for AI models on supercomputers is spent on networking and moving data between GPUs, not on compute.
  • Hyperscalers like Amazon and Google build custom chips to control costs, despite NVIDIA's CUDA software moat.
AI & Tech (2)
  • Whisperflow is a speech-to-text tool that outperforms others by fixing grammatical errors and allowing natural pauses during dictation.
  • AGI is a moving goalpost; technology that would have been considered AGI a decade ago is now seen as standard.
Labor (3)
  • A Quinnipiac poll shows 70% of Americans believe AI will decrease job opportunities, but only 30% are personally worried.
  • Jeremy Frankel argues AI automation is different because it automates cognition, not just physical labor, unlike past revolutions.
  • Victor Riparbelli is optimistic that future jobs will focus more on human enjoyment like dining and music, moving away from numerical work.
ModelsEnterpriseCodingBig Tech