Ep 2 · Full episode
Agents That Actually Ship
0:0020:40
Hosted by Rocco and Ruby. Follow the show for new episodes every Thursday.
Show notes
The demo is the easy ten percent. The other ninety — evals, guardrails, cost control, traceability, humans on exceptions — is what separates a viral video from a real product. Rocco and Ruby break down the shipping playbook with this week's receipts: Flexport's 21-million-task freight agents, Loadsmart's human-backed freight operators, Nvidia's silicon kill switch, and GPT-6.1 Sol's agent economics.
Segments
- The Build — The demo-to-production gap: Flexport MCP (21M tasks a year, custom eval platform, every action traced, humans on exceptions), Loadsmart's freight agents (~80% resolved hands-free, operators behind the rest), Nvidia's Open Agent Safety Platform (OpenShell + Sentry on BlueField-4, quarantine in milliseconds), and the GPT-6.1 Sol price reset (one-fifth the cost of Astra). The five-rule shipping playbook: trace everything, evals as gates, guardrails around money and motion, humans on exceptions, the kill switch is a feature.
- Loud Links — helplearnai.com: the hands-on AI school. Stop watching AI tutorials, start building with AI. The four prompt packs are available now.
- Hot Take — If your agent cannot fail safely, it cannot ship at all. Ungovernable means undeployed. Receipts: OpenAI held Astra back over authorization and reporting concerns while Sol shipped; Nvidia moved safety into silicon.
- Demo Derby — Flexport MCP freight booking, Loadsmart freight agents, GPT-6.1 Sol agentic coding at one-fifth the price, Nvidia OpenShell open-source sandbox — and the anti-demo of the week: rogue agent reports.
Links & references
- Flexport lets AI agents book, track, and optimize global freightbusinesswire.com →
- Loadsmart launches AI agents backed by freight operatorswcia.com →
- Nvidia launches platform for reining in rogue AI agentstechcrunch.com →
- OpenAI launches GPT-6.1 Sol at one-fifth the cost of Astratheoutpost.ai →
- GPT-6.1 Sol vs GPT-6 Astra: is the cheaper model actually better?ibtimes.sg →
- HelpLearnAI — this week's Loud Linkshelplearnai.com →
- Showcase — demos and projects on our radaraiextrovert.com →
Coming up next
Episode 3 — "The Prompt Is Not the Product" — why clever prompts don't survive contact with users, and what the teams turning LLM tricks into durable products do differently. Out now.