KORDUROYFriday, July 24, 2026 · Run 1 · 7:00 AM ET
Picking the Right Gemini Models, Building Design Systems & Auditing Agents
11 stories from 2 sources
TL;DR
- Gemini split into 3 models — pick Flash for everyday work, skip Flash Cyber unless you're gov/vetted.
- Google says workplace AI reaches 68% of U.S. occupations but only ~21% of tasks — useful adoption benchmark for client talks.
- OpenAI and Anthropic both shipped voice upgrades this week — a two-horse race is forming.
AI Tips3
Gemini is now 3 models — pick the right one
- Everyday coding & document analysis → Gemini 3.6 Flash (17% cheaper output tokens).
- High-volume, speed-critical jobs → 3.5 Flash-Lite (~350 tokens/sec).
- Flash Cyber is gov/vetted-partner only — don't build workflows around it.
- Access via Google AI Studio, the Gemini app, or Gemini Enterprise.
Source: The Rundown AI
Build a design system before generating slides
- Write a DESIGN.md: colors, fonts, spacing (e.g. your !m palette).
- Have Claude turn it into reusable slide components + a template.
- Save standing rules in a CLAUDE.md so future decks inherit your fixes.
Source: The Rundown AI
Audit an AI agent's decisions before you ship its work
- Before running a coding agent, lock the key decisions yourself (behavior, constraints, architecture).
- After it finishes, have it list every choice & assumption — review that list, not the whole code diff.
Before I merge: list every decision, assumption, and shortcut you made; rate your confidence in each; flag anything that only fixes this example rather than the general problem; and tell me honestly whether you'd stand behind this in production. Don't change any code yet — wait for my review.
Source: The Rundown AI
AI News4
Google's own data: workplace AI now reaches 68% of U.S. occupations — but touches only ~21% of tasks
Adoption is broad but shallow.
Why it mattersA realistic benchmark for client AI-adoption talks — frame impact at the task level, not wholesale job replacement.
Source: The Rundown AI
OpenAI is turning ChatGPT into a business platform
Adding ads, a small-business "agent" program, and a 3.2 GW data center in Georgia.
Why it mattersThe assistant is becoming an agent ecosystem worth tracking for client solutioning — and ads in ChatGPT may change how you position/trust its answers.
Source: The Rundown AI
OpenAI & Anthropic pull ahead as a "two-horse race" — both just shipped voice upgrades
Claude voice now runs on Opus/Sonnet with Gmail & Slack connected.
Why it mattersDirectly relevant to your tooling — Claude voice + connected apps could speed hands-free work, low cost to trial this week.
Source: The Rundown AI
OpenAI & Anthropic are lobbying Washington to restrict Chinese open-weight models
(Nvidia's Jensen Huang is pushing back).
Why it mattersInteresting industry politics — but unlikely to change your day-to-day work in the near term.
Source: The Rundown AI
Beyond AI4
Fed holds rates steady, signals no cut before September
Markets had priced in higher odds of an earlier move.
The takeawayClient budgeting conversations that assumed cheaper borrowing this quarter should be revisited.
Conversation starterAre we still planning around a rate cut this summer, or should we push that assumption out?
Source: Morning Brew
A major airline reported record summer bookings
Despite higher fares.
The takeawayNo direct relevance to your work, general awareness only.
Conversation starterAnyone else notice flights are packed and pricier this summer, but people are still booking?
Source: Morning Brew
A large retailer announced a round of corporate layoffs
Tied to AI-driven automation of back-office roles.
The takeawayDirectly relevant — a live example to reference when advising clients on AI-driven workforce planning.
Conversation starterHave you seen the layoffs tied to automating back-office work? Curious how leadership is framing it internally.
Source: Morning Brew
A tropical storm is tracking toward the Gulf Coast
Expected to make landfall this weekend.
The takeawayNo direct relevance to your work, general awareness only.
Conversation starterAnyone got clients or offices in the storm's path this weekend?
Source: 1440
Summarized from The Rundown AI, Morning Brew (Jul 23–24)
