Price:

AI & Tech

AI agents reshape businesses before they hit scale

Wednesday, May 27, 2026 · from 6 podcasts
  • Solo founders and Fortune 500 companies both use AI agents to strip out middle-management coordination costs.
  • Managed agent runtimes, not raw models, now drive AI performance and enterprise adoption.
  • The shift to AI-native workflows is eliminating traditional entry-level jobs in software and operations.

A solo founder uses Claude to solve a DNS configuration problem in five minutes, a task he would have outsourced two weeks ago. On a different continent, a Fortune 500 company builds a parallel digital twin of its invoice department using the same class of tools. These are not separate trends. AI agents are collapsing the distinction between startup agility and enterprise scale, making internal coordination more expensive than external execution.

Salim Ismail argues on Moonshots that Ronald Coase’s seminal theory of the firm is obsolete. The cost of holding a meeting to discuss a software feature now exceeds the cost of having an agent build it. This economic flip is why companies must stop trying to retrofit AI and instead build AI-native entities at their edge, migrating workflows one by one. Ismail estimates a successful transition yields 100x annual performance gains and lets companies eventually operate with a quarter of their current staff.

The primary casualty is middle management. Their core function - packaging information for leadership - is now a native AI capability. The C-suite’s role is shifting to oversight and exception handling, evaluating agent outputs rather than generating strategy themselves.

"Middle management will shrink by 60% as coordination tasks vanish, while the C-suite shifts to dashboard oversight and exception handling."

- Salim Ismail, Moonshots with Peter Diamandis

This isn't just about efficiency; it's about a new infrastructure layer. Nathaniel Whittemore notes on The AI Daily Brief that the industry has moved into a 'harness engineering' phase. A model’s performance is now dictated more by its runtime environment than its underlying weights. Data from Endor Labs shows GPT-5.5’s functionality score on a coding benchmark jumped from 61% to 87% when moved from its native harness into Cursor’s SDK.

These managed runtimes, or 'Harness as a Service,' abstract away the complexity of persistent memory and tool protocols. They allow non-developers to build agentic apps, which is accelerating adoption in mainstream industries. Jake Woodhouse cites a Melbourne construction company using Claude to analyze material costs and project timelines, collapsing days of manual work into minutes.

For developers, the landscape is polarizing. Milan, co-founder of NanoGPT, notes the market is splitting. Open-source models win on price and lack of censorship for automated tasks, but for high-stakes work like complex programming or medical advice, proprietary models like Claude 4.7 - used five times more than any other model on his platform - maintain a three-to-six-month intelligence lead.

The final barrier is organizational immune response. Ismail points out that 44% of Gen Z workers have sabotaged AI training to protect their jobs. The transition is less a technical challenge and more a cultural siege against a legacy system designed for human bottlenecks.

"We are moving past the era of 'hobbyist' agent building. Nathaniel Whittemore argues we have entered the age of 'Harness as a Service' (HaaS)."

- The AI Daily Brief: Artificial Intelligence News and Analysis

The wave is monetizing. Google Cloud revenue grew 63% year-over-year, with a $460 billion backlog, while Amazon is pouring capital into infrastructure spoken for by customer demand. The proof is shifting from speculative bubbles to booked revenue, funded by companies betting their survival on moving faster than their own meeting schedules.

Source Intelligence

- Deep dive into what was said in the episodes

The Organizational Singularity: AI-Proof Your Company | EP #258May 26

  • Ismail says companies must architect around intelligence instead of hierarchy, moving from human-centric workflows to AI-native, agentic systems.
  • The 'organizational singularity' centers on recursive self-improvement at the workflow level, enabling companies to learn faster than competitors.
  • Ismail warns any high-margin business line can be replicated by a small team using tools like OpenClaude in 60-90 days, making incumbent firms vulnerable.
  • Companies should build an AI-native digital twin at the edge, migrating workflows like invoice processing to it while leaving the legacy core untouched.
  • Ismail estimates a successful transition yields 100x performance improvements per year, and companies can eventually operate with 10-25% of their current workforce.
  • Middle management will shrink by 60% as coordination tasks vanish, while the C-suite shifts to dashboard oversight and exception handling.
  • Ismail's intelligence stack has six layers: purpose, sensing, interpretation, decision, orchestration, and learning, wrapped by a governance protocol for agent oversight.
  • Agent passports with metadata constraints prevent rogue actions, supported by granular rollback, searchable logs, and human review queues.
  • Ismail cites Cognitions Labs achieving 73x ARR growth after going fully AI-native, and says the full industry transition will take five to seven years.
  • Ismail's rewrite methodology involves backcasting from a future vision, scoring the company on seven dimensions like organizational drag and AI integration.
Also from this episode: (4)

AI & Tech (3)

  • Salim Ismail argues Coase's 1937 'The Nature of the Firm' model is obsolete because coordination costs inside a company now exceed external execution costs due to cheap AI agents.
  • Ismail notes 44% of Gen Z workers sabotage AI training to protect their jobs, exemplifying the organizational immune system that blocks change.
  • The new book 'Organizational Singularity' will be released as a Claude AI skill to stay updated, not as a static text.

Health (1)

  • Peter Diamandis highlights Fountain Life's full-body MRI and early cancer detection screening, noting 3.3% of members had undiagnosed cancers.

AI Sub Slave VS NanoGPT with Milan | FREEDOM TECH FRIDAY 41May 26

  • Milan explains NanoGPT offers subscription and pay-per-prompt access to hundreds of text, image, video, and audio AI models through a single interface, with strong privacy defaults and crypto payments.
  • Milan founded NanoGPT as a Telegram bot to let people pay small amounts in crypto to try ChatGPT, aiming to make AI accessible without credit cards or personal data after his own privacy experiences at a central bank.
  • Milan says NanoGPT prioritizes privacy by default, allowing use without accounts and storing data locally, but added optional encrypted storage and Google sign-in for users who want syncing.
  • Milan explains NanoGPT supports around twenty cryptocurrencies, with Monero being the most used for ten months, followed by Bitcoin and Nano, to meet users where they are and foster a circular crypto ecosystem.
  • Milan outlines NanoGPT pricing: a $12 monthly subscription for open-source models with a 30-60 million token weekly allowance, or pay-per-prompt starting at $1, where costs vary widely based on model and task complexity.
  • Milan notes users can optimize costs by using expensive models like Claude Opus as orchestrators and cheaper models for simple tasks, or by setting spending limits and choosing between premium, standard, or basic auto-model routing.
  • Milan says NanoGPT's OpenAI-compatible API allows integration with any coding harness or agent tool like Claude CLI, and its agents page provides endpoints for web search, image generation, and other functions discoverable by AI models.
Also from this episode: (11)

AI & Tech (10)

  • Milan says OpenAI's GPT 5.5 and Anthropic's Claude Opus 4.7 are the leading closed-source AI models for general chat and coding. Claude Opus 4.7 is used five times more than any other model on NanoGPT.
  • Milan states Google's Gemini 3.5 Flash is a top closed-source model because it is very fast and cheaper, while Opus 4.7 is expensive and slower.
  • Milan argues open-source text models like GLM 5.1, DeepSeek V4 Pro, and Kimi K 2.6 are three to six months behind the top closed-source models in quality for tasks like programming or medical advice.
  • Milan notes open-source competition drives down AI model costs and increases efficiency, while closed-source models have fixed prices due to monopolies from providers like OpenAI or Azure.
  • Milan observes running top AI models locally on consumer hardware like a MacBook or phone is currently unrealistic due to their size, but smaller image models are more feasible for local use.
  • Milan says proprietary image models like Midjourney and GPT Image 2 offer higher quality than open-source options, but they impose stricter content censorship than local fine-tuned models.
  • Milan describes users employing NanoGPT for creative applications like detailed role-playing in constructed worlds and automated polymarket betting operations where AI agents research and place wagers.
  • Milan states NanoGPT's development is driven 90% by user feedback, leading to additions like video and audio models, APIs, trusted execution environments, and a browser-installed local model for memory.
  • Milan demonstrates NanoGPT's local model feature, which downloads a 500 million parameter model directly in the browser for tasks like conversation memory, aiming for accessibility on basic devices.
  • Milan explains NanoGPT's global memory uses a local model to learn user background details across chats, while context memory expands a model's window by feeding it only the most important parts of long conversations.

AI Infrastructure (1)

  • Milan advises starting AI agent setups in a sandboxed environment like a separate computer or VPS and using AI models themselves to configure the workflows, rather than giving them blind access to emails or local files.

Behind the Scenes: Using AI to Build a Real Business in Real Time (JWP125)May 25

  • Jake Woodhouse sees AI adoption as more immediate than Bitcoin, noting AI already permeates daily life while understanding Bitcoin requires deeper financial inquiry.
  • Woodhouse runs an AI assessment product, a productized consulting service that interviews business owners and delivers bespoke reports recommending low-hanging fruit AI tools.
  • He cites a Deloitte report finding one in three Australian small business owners don't know where to start implementing AI.
  • Woodhouse uses Claude as a strategic and operational partner, employing a master thread for strategy and sub-threads for task execution.
  • He solved onboarding issues for Apollo, a cold email outreach tool, by screenshotting problems and sending them to Claude for step-by-step guidance.
  • Woodhouse built a lead magnet webpage on his existing WordPress site, creating a PDF and email capture form using ConvertKit, which he integrated and automated with Claude's help.
  • Claude assisted him in updating DNS records on Squarespace to ensure email deliverability, a task he would have previously outsourced.
  • He designed and launched a LinkedIn paid ad campaign targeting Australian accountants, using Claude to strategize and Canva to create ad imagery.
  • Woodhouse notes a friend's construction company uses a consultant to implement Claude for analyzing material costs and project management, drastically reducing time spent.
  • He argues AI assessments should target operational staff like COOs and project managers, not just CEOs, to create efficient workflows.
  • Woodhouse claims he built a complete marketing funnel from scratch in one week despite having no technical background and multiple other commitments.
Also from this episode: (2)

AI & Tech (2)

  • Woodhouse built a dedicated landing page for the LinkedIn ad traffic on WordPress, again using Claude for copy and design guidance.
  • He asserts AI tools enable execution two to four times faster than traditional methods like Google searches, podcasts, or business books.

#393 ‒ AMA #85: A guide to medications and supplements: determining what to take, what to skip, and how to know if they're working for youMay 25

Also from this episode: (6)

Longevity (5)

  • Peter Attia argues exercise is the single most effective lifespan and healthspan intervention, citing superior impact on all-cause mortality compared to smoking cessation, hypertension, lipid, and diabetes management.
  • Attia uses a 'Centenarian Decathlon' exercise with patients, forcing them to rank ten physical goals for their last decade and mapping the functional requirements needed to achieve them.
  • Attia's longevity strategy focuses on extending life without chronic disease, distinguishing it from the 'Medicine 2.0' model of managing life with chronic illness.
  • Attia states a zero coronary artery calcium score carries an approximate 15% risk of being a false negative, noting he has seen many cases of soft plaque on CTA scans after a zero CAC result.
  • Attia believes APOB is an unambiguous causal driver of atherosclerosis and must be treated even in metabolically healthy individuals, using an APOB of 150 as an example to treat down to a goal of 60 in a clean-artery patient.

Health (1)

  • Attia compares treating high APOB in a patient with pristine arteries to convincing a new smoker to quit before lung damage appears, arguing causality not certainty guides the intervention.

5/24/26: Neocons FREAK As Trump Says Iran Deal INCHES AwayMay 24

Also from this episode: (14)

Diplomacy (7)

  • A potential 60-day Iran-US deal is imminent, marked by a professionally vetted Trump Truth Social post that correctly names foreign leaders and shows prior consultation with regional partners.
  • Dr. Parsi notes a panic among Washington 'warmongers' despite the US employing sanctions, lethal strikes, harsh rhetoric, naval blockades, and weapons interdiction against Iran - all of which have failed.
  • Reported deal terms include a comprehensive cessation of hostilities extending to Lebanon, the gradual release of frozen Iranian assets, and an end to the US 'blockade of the blockade' in the Strait of Hormuz.
  • Maritime traffic through the Strait would resume under joint Iranian and Omani oversight, with a 30-day window to negotiate a final agreement addressing the nuclear issue and the Strait's long-term status.
  • The deal's first phase reverts to the original ceasefire terms before FDD advocated for the 'blockade of the blockade.' The real test comes in the second-phase talks on Iran's nuclear stockpile and the Strait.
  • Dr. Parsi criticizes the public 30-day negotiating window as a 'silver platter' for hawks to sabotage the deal and as making Trump politically vulnerable during that period.
  • Controlling the Strait of Hormuz is a new red line for Iran. A proposed Omani 'environmental management fee' is being discussed, with the US pushing for broader GCC regionalization to dilute Iranian control.

Politics (6)

  • Israeli public criticism of a deal may be muted during their election season because Trump remains extremely popular in Israel and could act as a domestic kingmaker.
  • Dr. Parsi argues military options to reverse US fortunes in Iran do not exist. Trump has shied from actualizing his threats because it would be a suicide mission that worsens the situation.
  • Krystal and Saagar highlight potential turning points for Trump: the initial Israeli intelligence failure, the disastrous Isfahan pilot rescue mission, and the unproductive China trip that closed off external pressure options.
  • The failed Isfahan operation, portrayed as a pilot rescue, likely involved an attempt to target nuclear facilities and resulted in the largest single loss of US carrier aircraft since the Vietnam War.
  • Ahead of Trump's Beijing visit, Iran and China struck a deal allowing Chinese ships through the Strait, neutralizing the FDD argument that China would pressure Iran and leaving the US as the only blocking power.
  • Dr. Parsi argues the war's outcome marks an inflection point ending US global primacy, raising fundamental questions worldwide about America's capacity and competence to sustain it.

War (1)

  • A final deal concluding this war would be a strategic defeat for Israel, revealing its inability to fight Iran without massive US support and further eroding its standing with the American public.

Why Agents Still Need HumansMay 24

  • Nathaniel Whittemore frames the agent landscape in three phases: the weights phase focused on model parameters, the context phase centered on prompts and RAG, and the current harness engineering phase, which builds persistent environments around static models.
  • Sam Altman told Ben Thompson the harness runtime is inseparable from model performance for effective agents, conceding he often cannot distinguish whether a great outcome stems from the model or its surrounding tools and state.
  • Google Cloud revenue grew 63% year-over-year, with a $460 billion backlog in new orders, up from $240 billion in Q4. CEO Sundar Pichai said AI is now the cloud unit's largest growth driver, though compute constraints limited revenue.
  • Google reported a 40% quarter-over-quarter surge in paid enterprise Gemini customers. The company's infrastructure now processes 16 billion tokens per minute, a 60% increase from the previous quarter.
  • AWS revenue grew 28% year-over-year, its fastest pace in nearly four years, making it a $152 billion annual recurring revenue business. Amazon added more server capacity than any other company in 2025.
  • Amazon's Q1 capital spending hit $43.2 billion, a 60% jump from last year, driving free cash flow down to $1.2 billion from nearly $26 billion. CEO Andy Jassy said the company's custom silicon business would be a top-three data center chipmaker if standalone.
  • Microsoft reported 39% Azure growth and 20 million paid Copilot seats, up from 15 million in January. CFO Amy Hood projected Azure's 40% growth rate to continue into Q2 and lifted annual CapEx guidance by $25 billion to $190 billion.
  • Meta reported quarterly revenue of $56.3 billion, up 33% year-over-year, but raised its 2025 CapEx forecast from $135 billion to $145 billion. The stock fell 5% as the market reacted negatively to the increased spending.
  • Whittemore argues harness as a service tools like the Cursor SDK represent a new infrastructure category, providing a pre-built agent runtime that handles tool dispatch, sandboxing, and error handling so builders only need to supply a model, tools, and a task.
  • An Endor Labs report found GPT-5.5's functionality score on a coding benchmark jumped from 61.5% to 87.2% when switched into Cursor's harness, demonstrating how the runtime environment dramatically changes model performance.
  • Early Cursor SDK use cases include a Gmail-integrated coding agent and a bug-catching agent that can view a live app in a browser, aiming to close the feedback loop between agent-written code and real-world performance.