Human In the Loop

Last week the US government gave Anthropic's best model a 72-hour public life, then switched it off for every foreign national on earth. This week, three Chinese labs shipped models that beat almost everything in the open, handed over the weights, and charged a tenth of the price. We figure out which of those two plays ends with the whole world building on your stack.

Welcome to Human In the Loop, a weekly podcast where two builders cut through the AI hype cycle. No breathless hype. No doomerism. No surface-level recaps.

This week:

SIGNAL OR NOISE

• FIFA is running the 2026 World Cup on AI. An Intelligent Command Center in Miami, Football AI Pro built with Lenovo on FIFA's own Football Language Model and handed to all 48 teams, 1,248 player avatars from one-second scans, and Gemini drawing up tactics for Argentina, Brazil, and France. The biggest live multi-model AI deployment on earth.

• China's open-weight wave. Z.ai's GLM-5.2 and Moonshot's Kimi K2.7 Code land with open weights and million-token windows, and Chinese labs now hold four of the top five open-weight spots at a fraction of Western pricing.

• Anthropic's Fable 5 and Mythos 5 are still dark a week after the US export ban, with no end date and a public defense from Anthropic.

• Noam Shazeer, co-author of the transformer paper, leaves Google for OpenAI two years after a reported $2.7B deal brought him back.

SHIP IT OR SKIP IT

Two real ideas, steelman versus pushback, no fence-sitting.

• An AI governance layer that keeps employee agents in their lane

• Vertical AI harnesses for one trade, like marketing or video

Closing hot takes:

→ Oscar: The most important AI lab of 2026 might not be American, and most US builders are too proud to notice.

→ Matt: A free model is only free if you own the GPUs, the ops team, and the eval harness to babysit it.

Your hosts:

• Oscar Gallo, AI Engineer and Entrepreneur. Builds AI products, ships code, runs companies.

• Matt Wozniak, Builder, operator, relentless executor. Builds, ships, scales.

Sources and full show notes are in the episode folder. Subscribe for a new episode every week. Comment with the AI term you keep hearing in meetings that nobody can actually define, and we will take it on next week.

#AI #WorldCup #FIFA #China #OpenWeights #Kimi #Anthropic #OpenAI #VerticalAI #AIAgents #AIGovernance #HumanInTheLoop

Anthropic shipped the most powerful public AI ever on June 9. Three days later the US government made them pull it from every foreign national on earth, including their own engineers.

This is Human In the Loop, a weekly podcast where two builders cut through the AI hype cycle. No hype. No doomerism. No filler.

SIGNAL OR NOISE

- Anthropic releases Fable 5 and Mythos 5 at half the old frontier price, five days after calling for a global AI pause

- The Trump administration puts both models under export controls after a Mythos jailbreak, and Anthropic disables access worldwide to comply

- OpenAI files confidentially for its IPO at a reported $730 to $850 billion, a week after Anthropic filed at $965 billion

- OpenAI lands inside Oracle Universal Credits, killing the enterprise procurement blocker

STACK CHECK: AI AGENTS IN PRODUCTION

The three layers to run an agent unattended.

- Claude Agent SDK, the agent loop you do not have to write

- LangGraph plus LangSmith, orchestration you can audit

- Temporal, durable execution so a crash is not a restart

HOT TAKES

Oscar: The Mythos ban is the best thing that ever happened to open weights. A closed model is one the government can switch off on a Friday.

Matt: Everyone watched the model and missed the money. OpenAI spent the week removing every reason a CFO can say no.

Hosts: Oscar Gallo (AI Engineer and Entrepreneur) and Matt Wozniak (Builder and Operator).

New episodes every week. Every fifth episode is a guest deep-dive.

The White House just signed an order that lets the government test the strongest AI models up to 30 days before you can. The labs said yes. Safety, or a backdoor? We make the call.

Human In the Loop is a weekly podcast where two builders cut through the AI hype cycle. No hype, no doomerism, no recaps.

SIGNAL OR NOISE

- Trump's frontier-model order, a 30-day government early-access window with the NSA deciding what qualifies

- GitHub Copilot flips to token billing, devs report 10x to 50x jumps

- Microsoft and Google crash the AI coding party with cheaper models

- ChatGPT crosses 1 billion users, and why Claude's 640% growth is the real story

- Anthropic calls for a global pause on frontier AI

NO JARGON REQUIRED

The three words on every AI invoice, explained for anyone who signs off on the bill.

1. Tokens, and why your bill went usage-based

2. Inference vs training, and what "inference-efficient" buys you

3. Evals and benchmarks, and why the leaderboard lies

HOT TAKES

Oscar: The executive order is a distribution deal. Every frontier release now has a 30-day government waiting room.

Matt: The Copilot billing meltdown is the best thing to happen to engineering leaders this year. The meter was always running.

Key links:

- White House AI order: https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/

- Copilot usage billing: https://github.blog/news-insights/company-news/github-copilot-is-moving-to-usage-based-billing/

- The token bill comes due: https://techcrunch.com/2026/06/05/the-token-bill-comes-due-inside-the-industry-scramble-to-manage-ais-runaway-costs/

Hosts: Oscar Gallo (AI Engineer & Entrepreneur) and Matt Wozniak (Builder, Operator). New episodes every week.

Six weeks ago Anthropic said a model was too dangerous to ship. This week they shipped Opus 4.8 and said Mythos-class models reach every customer soon. Was it ever that dangerous? We make the call.

Human In the Loop is a weekly podcast where two builders cut through the AI hype cycle. No hype, no doomerism, no recaps.

SIGNAL OR NOISE

- Claude Opus 4.8, near-Mythos alignment and Dynamic Workflows that run up to 1,000 subagents

- Anthropic passes OpenAI, a new round values it near $1 trillion

- Cognition raises $1B at $26B for Devin

- A poisoned Nx Console extension stole Claude Code configs and cloned 3,800 GitHub repos

- Pope Leo XIV's first encyclical takes on AI

SHIP IT OR SKIP IT

Three real ideas. One host pitches, the other pushes back. No fence-sitting.

1. The Company Brain (My First Million Ep 822)

2. Voice AI agent for the trades (Avoca's $125M raise)

3. Stripe for AI agents (Circle Agent Stack)

HOT TAKES

Oscar: "Too dangerous to ship" was never a safety call. It was a 90-day enterprise head start.

Matt: Devin at $26B is the top, not the floor. An agent lab with no distribution is the WeWork of 2026.

Key links:

- Opus 4.8: https://www.anthropic.com/news/claude-opus-4-8

- Anthropic tops OpenAI: https://www.cnbc.com/2026/05/28/anthropic-open-ai-startup-value.html

- CISA Nx breach: https://www.cisa.gov/news-events/alerts/2026/05/28/supply-chain-compromises-impact-nx-console-and-github-repositories

Hosts: Oscar Gallo (AI Engineer & Entrepreneur) and Matt Wozniak (Builder, Operator). New episodes every week.

Cursor shipped a coding model that matches Claude Opus 4.7 at one tenth the price. The same week, Anthropic is raising $30B at a $900B valuation. We make the call.

Human In the Loop is a weekly podcast where two builders cut through the AI hype cycle. No hype, no doomerism, no surface-level recaps.

SIGNAL OR NOISE

We run through the week's biggest AI headlines and call each one:

- Cursor Composer 2.5, top-three coding agent at $0.50 / $2.50 per million tokens

- Andrej Karpathy joins Anthropic's pre-training team

- Anthropic's $30B round at a $900B valuation

- Google I/O 2026: Gemini Spark agent and Antigravity 2.0

- Hark's $700M Series A, noise until something ships

STACK CHECK

Three dev tools we actually adopted in the last 30 days. What they replace, what they cost, would we keep using them:

1. Matt Pocock's `skills` repo, Skill Handoff (a real replacement for `/compact`)

2. Convex (convex.dev), TypeScript-native backend built for agentic workflows

3. Google Antigravity 2.0 + CLI + Managed Agents API

HOT TAKES

Oscar: Cursor Composer 2.5 just proved the open base model thesis. Coding-as-a-service is falling off a price cliff.

Matt: Karpathy joining Anthropic is worth more than the $900B valuation. Talent is the only moat the market still underprices.

Key links:

- Cursor Composer 2.5: https://cursor.com/blog/composer-2-5

- Karpathy at Anthropic: https://techcrunch.com/2026/05/19/openai-co-founder-andrej-karpathy-joins-anthropics-pre-training-team/

- Anthropic $900B round: https://www.bloomberg.com/news/articles/2026-05-12/anthropic-in-talks-to-raise-30-billion-at-900-billion-valuation

- Antigravity 2.0: https://techcrunch.com/2026/05/19/google-launches-antigravity-2-0-with-an-updated-desktop-app-and-cli-tool-at-io-2026/

Hosts: Oscar Gallo (AI Engineer & Entrepreneur) and Matt Wozniak (Builder, Operator). New episodes every week.

Three months ago we asked if there was a real business in selling AI to small businesses. This week Anthropic shipped it themselves. We watched the playbook get eaten by a frontier lab in 90 days.

Human In the Loop is a weekly podcast where two builders cut through the AI hype cycle. No hype. No doomerism. No surface-level recaps.

SIGNAL OR NOISE

The week's biggest AI headlines, called:

- Anthropic ships Claude for Small Business (15 workflows + QuickBooks/HubSpot/Microsoft 365 connectors + 10-city US tour)

- Thinking Machines previews Interaction Models (276B MoE, 0.4s full-duplex response)

- Sierra raises $950M at $15B+ ($150M ARR in 8 quarters, 40% of Fortune 50)

- Trump admin reverses on AI oversight (national-security fears post-Mythos)

- Cloudflare cuts 1,100 jobs (~20%) and blames AI — same day it posts record Q1 revenue

NO JARGON REQUIRED

The three AI concepts every exec and operator keeps hearing this quarter, in plain English:

1. "AI Agent" — the word every vendor abuses (real agents vs chatbots in costume)

2. Connectors and MCP — what "AI plugged into your tools" actually requires

3. Full-Duplex Voice (Interaction Models) — what to ask when a vendor pitches a voice agent

For each: plain-English explanation, plus a decision framework. Invest, wait, build, buy, or ignore.

HOT TAKES

Oscar: Anthropic just turned every AI consultant into a reseller. The workflow layer is going to the model providers.

Matt: Every vendor calling their thing an "AI agent" in 2026 is going to look like every vendor that called their thing "cloud-native" in 2015.

Key links:

- Claude for Small Business: https://www.anthropic.com/news/claude-for-small-business

- Sierra $950M raise: https://techcrunch.com/2026/05/04/sierra-raises-950m-as-the-race-to-own-enterprise-ai-gets-serious/

- Model Context Protocol: https://www.anthropic.com/news/model-context-protocol

- Thinking Machines Interaction Models: https://techcrunch.com/2026/05/11/thinking-machines-wants-to-build-an-ai-that-actually-listens-while-it-talks/

Do you want to know more? https://podcast.vallyseed.com/Wanna be part of that of the podcast? [email protected]

Hosts: Oscar Gallo (AI Engineer & Entrepreneur) and Matt Wozniak (Builder, Operator).

New episodes every week.

Your token bill is going up 4–8x in the next two years. Meanwhile, there's a profitable AI company running production workloads for SMBs on Mac Minis — eliminating token costs entirely and serving 600–800K API calls a month on a single machine.

This is a guest deep-dive — every 5th episode we drop the segments and go one topic, one expert. This week: Jackson Oaks, founder of Recursion AI and the self-hosted AI platform Courier.

Courier: https://getcourier.ai

Jackson on LinkedIn: https://www.linkedin.com/in/jackson-oaks/

THE CONVERSATION

- The 80/20 of open-source AI — why 80% of businesses have 80% of use cases that don't need frontier models

- Why Apple M-series is the most underrated AI hardware story of the decade (1/10th the cost of NVIDIA, 30–40x more power efficient)

- Real production numbers: 600–800K API calls/month per Mac Studio, flat-rate pricing, the math behind killing token spend

- A Fortune company paid Deloitte $400K for a "GPT wrapper with a RAG database"

- Why MIT found 95% of business AI pilots couldn't measure ROI

- War stories: the infinite hallucination loop that ate 10K requests in a week, and how they built hallucination detection from scratch

- Fine-tuning a 14B model that outperformed GPT-4o on a structured extraction task

HOT TAKES

Matt: Anthropic and OpenAI consolidate within three years — or open-source overtakes both and they shrink dramatically.

Oscar: Every home will run a better Alexa on a Mac Mini powered by open-source models.

Jackson: Better systems with smaller models beat throwing frontier models at unstructured problems.

Hosts: Oscar Gallo (AI Engineer & Entrepreneur) and Matt Wozniak (Builder & Operator).

Guest: Jackson Oaks, Recursion AI / Courier.

New episodes every week. Every 5th episode is a guest deep-dive.

On April 24th, an AI coding agent deleted a production database in 9 seconds, then confessed: "I violated every principle I was given." Five days later, the lab behind that model took funding offers at a $900 billion valuation. We call every story.

Human In the Loop is a weekly podcast where two builders cut through the AI hype cycle. No hype, no doomerism, no surface-level recaps.

SIGNAL OR NOISE

We run through the week's biggest AI headlines and call each one:

- Anthropic eyes a $900B valuation in a $50B raise

- AI agent deletes PocketOS production database in 9 seconds

- China blocks Meta's $2B Manus deal, Meta buys Assured Robot Intelligence days later

- Cloudflare and Stripe ship the agent provisioning protocol

- Chinese court rules an AI-replacement layoff was illegal (NPR, May 1)

SHIP IT OR SKIP IT

Three real ideas pulled directly from Y Combinator's Summer 2026 RFS. Steelman, pushback, verdict. No fence-sitting.

1. AI-Native Service Companies (RFS by Gustaf Alströmer)

2. Company Brain (RFS by Tom Blomfield)

3. Counter-Swarm Defense (RFS by Tyler Bosmeny)

HOT TAKES

Oscar: Anthropic at $900B is sovereign wealth fund pricing. Treat your model spend like an interest rate.

Matt: Every AI agent in production is one bad scope away from being PocketOS. The fix is permissions and audit logs, not better models.

Key links:

- Anthropic $900B: https://www.bloomberg.com/news/articles/2026-04-29/anthropic-considering-funding-offers-at-over-900-billion-value

- AI deletes database: https://www.theregister.com/2026/04/27/cursoropus_agent_snuffs_out_pocketos/

- Cloudflare agent protocol: https://blog.cloudflare.com/agents-stripe-projects/

- NPR, Chinese court rules AI-replacement layoff illegal: https://www.npr.org/2026/05/01/nx-s1-5807131/tech-worker-china-ai

Hosts: Oscar Gallo (AI Engineer and Entrepreneur) and Matt Wozniak (Builder, Operator).

New episodes every week.

Two weeks ago we called GPT-5.5 noise. It shipped Thursday. It beat Claude Mythos on Terminal-Bench. Google bet forty billion on Anthropic the same day. The race got weirder.

Human In the Loop is a weekly podcast where two builders cut through the AI hype cycle. No hype, no doomerism, no surface-level recaps.

SIGNAL OR NOISE

We run through the week's biggest AI headlines and call each one:

- GPT-5.5 (Spud) ships and beats Claude Mythos on Terminal-Bench 2.0

- DeepSeek V4 ships open source on Huawei chips, 1M context, one-sixth the cost

- Google invests up to $40B in Anthropic

- Cognition (Devin) in talks at $25B valuation

- Grok 4.3 Beta at $300 per month, still no cross-session memory

STACK CHECK

Four tools we actually use right now. What stuck, what didn't, and whether we'd keep paying next month.

1. Vercel Labs Agent Browser

2. WorkOS (FGA + AuthKit for agents)

3. AI SDK (ai-sdk.dev)

4. Claude Code /remote-control

HOT TAKES

Oscar: Open-source AI just won the cost war and lost the trust war the same week.

Matt: The "big release" era is over. From now on, model launches are software updates.

Key links:

- Introducing GPT-5.5: https://openai.com/index/introducing-gpt-5-5/

- DeepSeek V4: https://techcrunch.com/2026/04/24/deepseek-previews-new-ai-model-that-closes-the-gap-with-frontier-models/

- Google + Anthropic $40B: https://www.cnbc.com/2026/04/24/google-to-invest-up-to-40-billion-in-anthropic-as-search-giant-spreads-its-ai-bets.html

- AI SDK: https://ai-sdk.dev/docs/introduction

- Claude Code Remote Control: https://code.claude.com/docs/en/remote-control

Hosts: Oscar Gallo (AI Engineer and Entrepreneur) and Matt Wozniak (Builder, Operator).

New episodes every week.

Anthropic shipped their best public model this week. In the same breath, they said it's their second best. The best one is gated to 50 enterprises.

Human In the Loop is a weekly podcast where two builders cut through the AI hype cycle. No hype, no doomerism, no surface-level recaps.

SIGNAL OR NOISE

The week's AI headlines, filtered:

- Claude Opus 4.7 ships with a 1M context window at standard pricing

- OpenAI's updated Agents SDK (sandboxed agents, Python first)

- AI on the desktop (Google's Gemini Mac app + Perplexity Personal Computer)

- Stanford's 2026 AI Index: 10% of Americans are more excited than concerned vs. 56% of experts

- Allbirds pivots from sneakers to AI, stock up 582% then crashing

NO JARGON REQUIRED

Three AI concepts decision-makers need to make real calls on this quarter. Plain English, then a decision framework.

1. Context Engineering. The successor to prompt engineering.

2. Agentic AI. What it actually is, and why Gartner says 40% of projects will be canceled by 2027.

3. Frontier Model Tiering. Why the best model is one you probably cannot buy.

HOT TAKES

Oscar: Context engineering is "write good internal documentation" rebranded so consultants can charge for it.

Matt: Every executive asking "what's our agentic AI strategy" is two years from being replaced by one who already shipped something.

Key links:

- Opus 4.7 release: https://platform.claude.com/docs/en/about-claude/models/whats-new-claude-4-7

- OpenAI Agents SDK: https://openai.com/index/the-next-evolution-of-the-agents-sdk/

- Stanford AI Index: https://spectrum.ieee.org/state-of-ai-index-2026

- Gartner forecast: https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027

Hosts: Oscar Gallo (AI Engineer & Entrepreneur) and Matt Wozniak (Builder, Operator).

New episodes every week.