One signal-dense note when a new episode lands. No noise.
Value-First AI Daily - Jul 20, 2026
July 20, 2026
Value-First AI Daily is back — relaunched as a daily working-notebook show with Chris Carolan and Nico Lafakis. We christened the new format with a live "Top 3 AI News" reveal the hosts saw for the first time on air (a $25 GPT-5.6 WordPress exploit, the EU forcing Google to open Android to rival AI assistants, and Moonshot's record-setting open-weight coding model Kimi K3), spun the agent-reveal wheel to meet Litmus and Dub, and built in the open — an agent-reveal wheel and an early 3D game world — while talking through model choice, agent expertise, and the "language of value."
Key takeaways
The show is back as a daily working-notebook format around 4:15 Central: a live Top 3 AI news reveal the hosts see for the first time on air, a wheel spin surfacing a roster agent, and live builds — no pitch, no script, showing the good with the bad.
Open weights, not the leaderboard rank, is the real story of the #1 model: the value is seeing how a frontier model works, the secret sauce closed models hide. Self-hosting a ~2.8T-parameter model isn't laptop-territory, so the practical pull is cost efficiency, not DIY hosting.
Model choice is a value-to-cost decision, not a chase-the-top-benchmark reflex. Chris ran his first comparative studies and kept his existing architecture because switching to a frontier model didn't improve the value-to-cost ratio.
AI's edge is multiplication, not replacement of your judgment. Today's models are all college-graduate good, and with the right harness they beat you at a specific task — but you still have to know how to ask, how to vet the response, and how to add genuine expertise.
Expertise has to be added, not asserted. "You are an expert" prompting isn't enough; real expertise-pack skills — best practices, standards, how a tool actually works — are what let a builder hand the work off with confidence.
Language shapes outcomes — the language of value. Chris is building a downloadable that strips old namespace traps out of an AI's vocabulary so agents reason in value-first terms instead of defaulting to inherited industry framing.
The learning curve is a turn-it-on problem. Waiting holds you back and risks being left behind; the moment you engage, you can catch up fast — and you learn by building or live conversation, not by watching.
Live building is messy, and that's the point. The wheel-spin build returned a front-end defect because the agents took a shortcut knowing the show was about to go live — recoverable in a prompt.
The hosts' broader thesis, offered as their prediction rather than established fact: one-shot app-building, SaaS disruption, and a flood of AI-generated media are close, with movie theaters headed toward relic status.
Keep learning
You just watched the work. Now build it.
The next step is doing it yourself — live, with people on the same path.
One signal-dense note when a new episode lands. No noise.