Synthesis

Agentic engineering

Building software by directing fleets of coding agents rather than writing code by hand, up to the 'dark software factory' where humans neither write nor review code.

Type
Synthesis
Confidence
Medium
Sources
17 videos
First published
Updated

Across 20 episodes the hosts, mainly Dave Morin and Sam Lessin, describe building as a new skill: you speak to the model, not the compiler. Dave defines a dark software factory as an organisation where code is neither written nor reviewed by humans, citing xAI’s reorg and Anthropic as moving that way (his report). Examples are personal: Dave built an iOS CRM for Offline in about two hours for roughly $150 of tokens (his figure), Jess builds an app a day and a headline-scoring tool by email, Sam gave an agent a spare Mac Mini to ship iOS apps and runs about $3,000 a month of always-on infrastructure. Dave says OpenClaw’s core engineers run 50 to 100 agents at once, limited by RAM and Apple silicon; he works on OpenClaw and has a stake. Sam describes the Ralph loop (a long PRD worked through in fresh contexts) but has not tried it. Dave and Sam both argue non-engineers are not too late, and Sam urges building to learn instead of debating. Guests with a product to sell (Suds on a CI company, Guillermo Rauch of Vercel on agent-driven growth) say agents move the bottleneck to CI and infrastructure. Dave prefers CLIs to MCP. Evidence is self-reported by two heavy users.

Key points

  • A dark software factory: agents write and review all code, humans only seed the system (Dave’s description) — 2 episodes. (Dave · Why OpenAI Can’t Win The Decen (Feb 2026) ▶ 46:35)
  • Non-engineers are not too late; agentic engineering is a new skill and the tools improved sharply in three months — 2 episodes. (Dave · Every CTO Is Quietly Switching (Jul 2026) ▶ 29:36)
  • Hosts build real tools by talking to agents: an iOS CRM in two hours, an app a day, a headline scorer by email, shippable products in 24 hours at a bot school — 4 episodes. (Dave · Agentic AI Will Break the Inte (Jan 2026) ▶ 08:48)
  • Run many agents in parallel; the constraints are tokens, local memory and your own supervision — 5 episodes. (Dave · Google I/O Just Killed SEO? Ag (May 2026) ▶ 36:27)
  • Long PRD-driven loops with a fresh context per task (Ralph), described by Sam and Brit but untried by Sam — 1 episode. (Sam · The Seed Market Is Broken: Meg (Jan 2026) ▶ 20:33)
  • Bottlenecks move: more code means CI and infrastructure strain, per two guests selling in that space — 2 episodes. (Suds · Suds (Sep 2026) ▶ 42:39) (Guillermo Rauch · Guillermo Rauch (Apr 2026) ▶ 08:38)
  • Stop arguing about product ideas in the abstract; build a version and learn — 1 episode. (Sam · Every CTO Is Quietly Switching (Jul 2026) ▶ 15:09)
  • A harness (software around a model) is the unit of work, and CLIs beat MCP (Dave) — 5 episodes. (Dave · Why the Met Gala Hates Tech, E (May 2026) ▶ 32:14) (Guillermo Rauch · Guillermo Rauch (Apr 2026) ▶ 21:01)

Frameworks

  • Dark software factory — Dave’s description of an agentic engineering organisation in which agents write and review all code and humans only seed the system and write the tests it is validated against. (Dave · Why OpenAI Can’t Win The Decen (Feb 2026) ▶ 46:35)
    • Code must not be written by humans
    • Code must not be reviewed by humans
    • Humans build tools to understand what is being built and to keep throughput high
    • Humans seed the initial intent, then validate against the tests they write

Tactics

  • Run a long PRD-driven loop with a fresh context window per task (Ralph) — Write a PRD with the model; the model works each task in a new context window until finished, then moves to the next for 6 to 7 hours. (Sam · The Seed Market Is Broken: Meg (Jan 2026) ▶ 20:43)
  • Give an agent full control of a spare Mac Mini to ship iOS apps — Sam gave his agent full control of a Mac Mini so it deploys apps through TestFlight from single commands. (Sam · Suds (Sep 2026) ▶ 07:43)
  • Have the agent build an internal mobile app in spare moments. — Dave booted a new iOS project at halftime of his son’s basketball game and talked to the agent through the day; the app was done in about 2 hours and went on colleagues’ phones. (Dave · Agentic AI Will Break the Inte (Jan 2026) ▶ 08:28)
  • Give a multi-agent engineering team enough local RAM and Apple silicon — OpenClaw worked with Apple on Mac Studios in custom configuration so engineers can run many agents at once. (Dave · Meta’s AI Campaign, While Riva (Jul 2026) ▶ 35:06)
  • Run many agents at once to drive token usage and throughput — Dave says token spend climbs when you run north of 10 agents at once; core engineers run 50 to 100 agents with continuous integration. (Dave · Google I/O Just Killed SEO? Ag (May 2026) ▶ 36:27)
  • Email your agent to build an internal tool and iterate on it live in a meeting — Jess emailed her bot, trained on The Information’s history, to build an app suggesting three headlines per story and scoring a typed headline out of five, red at three or below; she demoed it to editors that day and raised options from three to six while in the meeting. Editors can override it. (Jess · Meta’s $17B Reckoning & Why Pa (Aug 2026) ▶ 05:26)
  • Spend an hour a day learning agentic tools if you have another job. — Dave: take an hour a day, and keep practising like learning guitar if you have another job; Sam counters elsewhere that he would binge it. (Dave · Every CTO Is Quietly Switching (Jul 2026) ▶ 32:51)

Stories & examples

  • Offline CRM iOS app in two hours: Dave had always wanted an iPhone app for Offline’s custom internal CRM; with the agent he built it in about 2 hours over a weekend for about $150 of tokens, and colleagues now use it. Lesson: Compare agent token cost with hiring an engineer (tens of thousands of dollars and months). (Dave · Agentic AI Will Break the Inte (Jan 2026) ▶ 08:48)
  • Sam’s bot school for influencers: Sam says he sat social media influencers with big followings down with a tool chain, including DigitalOcean, and in 24 hours they shipped products to real users, versus hackathons that used to yield half-working demos in four days. Lesson: The most uplifting part of AI is the empowerment of tinkerers. (Sam · Every CTO Is Quietly Switching (Jul 2026) ▶ 30:31)
  • A founder who thought he was too late: Dave says he and Brit spoke yesterday with a founder who spent years at Apple and Spotify working with engineers and believed he lacked the skill set for agentic engineering; Dave told him he was more capable than he thought. Lesson: People who are not engineers underestimate their ability to work with agentic tools. (Dave · Every CTO Is Quietly Switching (Jul 2026) ▶ 29:36)
  • Jess’s headline-scoring app: Jess built an app by email in one go: three headline suggestions per story and a one to five score, demoed to editors in a 9:15 meeting and updated live. Lesson: Agents let a non-engineer ship an internal tool in minutes; policy is that editors override. (Jess · Meta’s $17B Reckoning & Why Pa (Aug 2026) ▶ 05:47)
  • Astra and a 12-year-old’s 3D model: Dave’s son used Astra to make a 3D model in an hour that was already printing on his 3D printer; Dave says the speed made him understand the letter’s warnings. Lesson: Models good at using computers, run at the scale of hundreds of thousands, are what make the safety debate real. (Dave · iPhone Duo, Meta Muse & Why AI (Sep 2026) ▶ 34:15)
  • CI company for 50x more code: Suds incubated a company aimed at the CI bottleneck created when engineers ship far more code. Lesson: AI-generated code volume breaks existing shipping pipelines. (Suds · Suds (Sep 2026) ▶ 42:49)

Numbers

  • Time to build the Offline iOS CRM app: 2 hours (host’s framing, approx.) (Dave · Agentic AI Will Break the Inte (Jan 2026) ▶ 08:34)
  • Sam’s bot infrastructure cost: 3000 dollars per month (host’s framing, approx.) (Sam · Peter Deng & Rachel Holt (Aug 2026) ▶ 06:22)
  • Agents run at once by OpenClaw core engineers: 100 agents (up to 50 to 100) (host’s framing, approx.) (Dave · Google I/O Just Killed SEO? Ag (May 2026) ▶ 36:34)
  • Increase in code shipped per feature: 50 times (x) (self-reported, approx.) (Suds · Suds (Sep 2026) ▶ 42:49)
  • Hours for bot-school participants to produce shippable products: 24 hours (host’s framing) (Sam · Every CTO Is Quietly Switching (Jul 2026) ▶ 30:42)
  • Increase in infrastructure demand from agents: 100 times (self-reported, approx.) (Guillermo Rauch · Guillermo Rauch (Apr 2026) ▶ 08:38)

Caveats & warnings

  • Suds cannot fully explain the CI company he promotes (guest promotes own company). (Suds · Suds (Sep 2026) ▶ 43:56)
  • Sam has not tried Ralph himself. (Sam · The Seed Market Is Broken: Meg (Jan 2026) ▶ 21:01)
  • Sam says ‘I’m too late’ is a mistake he has continuously made even in investing, which his planned book chapter ‘you’re not too late’ addresses. (Sam · Every CTO Is Quietly Switching (Jul 2026) ▶ 35:12)
  • Dave discloses he is speaking his own book because he works on OpenClaw. (Dave · Meta’s AI Campaign, While Riva (Jul 2026) ▶ 11:58)

How much to trust this

  • Evidence: 20 episodes, 3 independent speakers (the four hosts and unidentified host speech count as one voice, the show; guests counted individually).

Sources

  1. (01)

    Google's AI-First Laptop, Meta's Spy Games, AI Monks in Middle America

    YouTube

    42:14 into “Google's AI-First Laptop, Meta's Spy Games, AI Monks in Middle America”

  2. (02)

    Is Google About to Kill Canva? │Claude Goes Rogue, Tesla vs. Waymo

    YouTube

    7:43 into “Is Google About to Kill Canva? │Claude Goes Rogue, Tesla vs. Waymo”

  3. (03)

    Why the Met Gala Hates Tech, Elon vs OpenAI Drama, and the Rise of Chinese AI Models

    YouTube

    32:14 into “Why the Met Gala Hates Tech, Elon vs OpenAI Drama, and the Rise of Chinese AI Models”

  4. (04)

    OpenClaw vs Meta vs OpenAI: The Personal Agent Wars Heat Up

    YouTube

    10:56 into “OpenClaw vs Meta vs OpenAI: The Personal Agent Wars Heat Up”

  5. (05)

    iPhone Duo, Meta Muse & Why AI Finally Scared Dave Morin

    YouTube

    34:15 into “iPhone Duo, Meta Muse & Why AI Finally Scared Dave Morin”

  6. (06)

    Meta’s $17B Reckoning & Why Parents Can’t Compete With Big Tech's Algorithm

    YouTube

    5:26 into “Meta’s $17B Reckoning & Why Parents Can’t Compete With Big Tech's Algorithm”

  7. (07)

    The Seed Market Is Broken: Mega-Rounds, Narrative Games, and Walkouts

    YouTube

    20:26 into “The Seed Market Is Broken: Mega-Rounds, Narrative Games, and Walkouts”

  8. (08)

    AI Can Change Everything and Still Be a Terrible Business + Peptide 101 from Matt Mazzeo

    YouTube

    37:39 into “AI Can Change Everything and Still Be a Terrible Business + Peptide 101 from Matt Mazzeo”

  9. (09)

    Why OpenAI Can't Win The Decentralized AI Future (OpenClaw, Apple's Win, X.AI Exodus)

    YouTube

    45:54 into “Why OpenAI Can't Win The Decentralized AI Future (OpenClaw, Apple's Win, X.AI Exodus)”

  10. (10)

    Agentic AI Will Break the Internet (ClawdBot, OpenAI & the AI Demand Delusion)

    YouTube

    8:28 into “Agentic AI Will Break the Internet (ClawdBot, OpenAI & the AI Demand Delusion)”

  11. (11)

    Meta's AI Campaign, While Rivals Ask Washington To Slow it Down | Lilian Weng Saga, China Cracks DUV

    YouTube

    11:58 into “Meta's AI Campaign, While Rivals Ask Washington To Slow it Down | Lilian Weng Saga, China Cracks DUV”

  12. (12)

    Fire Sam Altman, The End of Software Engineers, and Why AI Is All Narrative

    YouTube

    30:53 into “Fire Sam Altman, The End of Software Engineers, and Why AI Is All Narrative”

  13. (13)

    The Risk Of Securitizing GPUs Like Long-Lived Infrastructure & More - A MOL Special Team Edition

    YouTube

    6:14 into “The Risk Of Securitizing GPUs Like Long-Lived Infrastructure & More - A MOL Special Team Edition”

  14. (14)

    Google I/O Just Killed SEO? Agentic Search, Universal Cart, and the New Web

    YouTube

    36:27 into “Google I/O Just Killed SEO? Agentic Search, Universal Cart, and the New Web”

  15. (15)

    Vercel CEO: 70% of Our Traffic Is Now AI Agents "Nobody Was Prepared" | Anthropic, OpenClaw, OpenAI

    YouTube

    8:38 into “Vercel CEO: 70% of Our Traffic Is Now AI Agents "Nobody Was Prepared" | Anthropic, OpenClaw, OpenAI”

  16. (16)

    Anthropic’s Bet on Coding Is Working (OpenAI Shopping Pivot, A16Z’s Top 50 List, $1B Tennis Channel)

    YouTube

    33:49 into “Anthropic’s Bet on Coding Is Working (OpenAI Shopping Pivot, A16Z’s Top 50 List, $1B Tennis Channel)”

  17. (17)

    Every CTO Is Quietly Switching Off Anthropic and OpenAI | GLM 5.2, Vox Americanus, Claude Fable 5

    YouTube

    15:09 into “Every CTO Is Quietly Switching Off Anthropic and OpenAI | GLM 5.2, Vox Americanus, Claude Fable 5”

(By email)

Compounding Insights