# How to Get the Most from AI This Summer *The AI Daily Brief — Sunday, 2026-07-26 · https://aidailybrief.ai/e/2026-07-26* **The skill now is managing agents, not chatting with a bot.** Ethan Mollick's updated guide draws a bright line between two interaction patterns: the old back-and-forth chat, where almost any model will do, and the new agentic workflow, where you delegate hours of work to a system that plans and acts on its own. That second pattern isn't just a new tool — it requires reprogramming how you think about AI. AIDB's free Summer Adventure is built to help you close your own "capability overhang": the gap between what AI can already do and what you're actually using it for. --- ## By the numbers - **10 min** — Agents did what would've taken hours prepping Mollick's MBA seminar - **195** — References the AI chased down in a 30-minute check of Mollick's book — no hallucinations - **20+** — Destinations (labs/projects) in the AI Summer Adventure - **10** — Stamps to reach "globetrotter" status - **150-300** — Words in the paste-ready "global ID" block from the Pack Your ID project - **2** — Real choices for intensive work, per Mollick: ChatGPT or Claude ## Main episode ### Using AI now means two very different things `[01:00]` Mollick's new guide splits AI use into two categories: the old chatbot back-and-forth, and agentic systems that can do the equivalent of many hours of human work in one go by pairing a model's brains with tools that let it plan and act. As he puts it, an agentic system basically gives an AI a computer to use. *For: Ops, Exec* Link: https://aidailybrief.ai/e/2026-07-26#two-interaction-patterns ### For low-stakes answers, any model will do `[01:00]` For quick, low-stakes tasks like asking for a recipe or help drafting a letter, Mollick says nearly any model — including the free ones — is good enough. For anything you really care about, use a premier model set to a high thinking level. *For: Ops* Link: https://aidailybrief.ai/e/2026-07-26#low-vs-high-stakes ### Only two real choices for serious work: ChatGPT or Claude `[01:00]` When it comes to intensive agentic work, Mollick argues there are effectively only two options right now — ChatGPT or Claude. Notably, that means Gemini is officially out of his rankings, at least for now. *For: Eng, Ops* Link: https://aidailybrief.ai/e/2026-07-26#only-two-choices ### Google… has fallen behind where it now counts. `[05:00]` *— Ethan Mollick, "An opinionated guide to which AI to use to do stuff"* Mollick writes that Google, which led on benchmarks not long ago, has no leading frontier model and nothing close to Codex or Claude Code — which is why he doesn't suggest Gemini as a primary system, though he notes that could change quickly. Link: https://aidailybrief.ai/e/2026-07-26#gemini-fallen-behind ### The real unlock is connecting AI to your stuff `[02:00]` Mollick's nudge is to actually use connectors — he's wired his systems into email, a non-private Google Drive, and many other apps. The "harness" he's describing is the set of controls and permissions that shape what the models can do, without ever using the word. *For: Ops, Eng* Link: https://aidailybrief.ai/e/2026-07-26#give-ai-a-computer ### "Prep for my seminar" — and the agents just went to work `[03:00]` Mollick told both ChatGPT and Claude to connect to Gmail and prep his MBA seminar, including building demos and answering messages. Both correctly figured out the target Monday was in September, did web research, built teaching materials, and drafted responses — returning results in about 10 minutes that would've taken hours. *For: Ops, Product* Link: https://aidailybrief.ai/e/2026-07-26#mba-seminar-demo ### Permissions matter a lot — one agent sent the email `[03:00]` In the seminar test, Claude only drafted a reply while ChatGPT actually sent an email to a colleague — because Mollick had previously granted it send permission. His lesson: until you trust a system and understand its mistakes, leave everything set to ask for approval first. *For: Ops, Legal, Exec* Link: https://aidailybrief.ai/e/2026-07-26#permissions-matter ### Computer use lets AI literally take over your machine `[04:00]` With computer-use turned on in Claude Code or Codex, the AI can control your mouse, browser, and computer. Mollick had it download Blender — a program he doesn't know how to use — and model an otter using a laptop on an airplane. He flags it as a security concern to approach carefully. *For: Eng, Product* Link: https://aidailybrief.ai/e/2026-07-26#computer-use-blender ### 195 references checked, zero hallucinations `[05:00]` Mollick fed the full PDF of his professionally edited upcoming book to the AI, which worked 30 minutes, chased down 195 references, and returned pages of accurate notes — no invented page numbers, no fake text. His only complaint: it was too nitpicky, which he overrode with human judgment. *For: Legal, Ops* Link: https://aidailybrief.ai/e/2026-07-26#book-check-195-refs ### Working with these systems is more like managing than it is chatting. `[05:00]` *— Ethan Mollick* Mollick's framing: you can almost think of AI agents as a team you delegate work to. NLW's read is that this is the core reprogramming required — a big shift in working, not just a new set of skills. *For: Exec, Ops* Link: https://aidailybrief.ai/e/2026-07-26#managing-not-chatting ### Google isn't useless — just not your main driver `[05:00]` Even while dropping Gemini from his primary recommendations, Mollick points to Google's Gemini Notebook (formerly NotebookLM) and richer media models like a video-capable Gemini model as things worth using. *For: Marketing, Ops* Link: https://aidailybrief.ai/e/2026-07-26#google-notebook-media ### Mollick is making agent-management feel approachable `[06:00]` NLW's take: rather than going model-by-model, Mollick draws a bright line between the old chat pattern and the new managing-agents pattern — deliberately making agent management feel accessible to people whose eyes glaze over at words like "harness." It's simple on purpose, because he's writing for a far less mature AI audience. *For: Exec, Ops* Link: https://aidailybrief.ai/e/2026-07-26#bright-line-take ### AIDB launches a free "Summer Adventure" training `[07:00]` The new self-directed program at summeradventure.ai is a choose-your-own-adventure to expand your AI skills, following earlier AIDB programs like New Year, Claw Camp, and Agent OS. It's organized around 20+ travel-themed "destinations," each a lab or project teaching a new skill. *For: Ops, HR* Link: https://aidailybrief.ai/e/2026-07-26#summer-adventure-launch ### Beginner to advanced, filterable by topic `[09:00]` Destinations span beginner, intermediate, and advanced, filterable by level or by topic — tools, agents and automation, building and creating, or setting up knowledge and context. Complete a project and you "stamp your passport," which can stay private or be shared publicly. *For: Ops* Link: https://aidailybrief.ai/e/2026-07-26#filter-by-level ### From day tripper to globetrotter `[09:00]` The gamified tiers: one stamp makes you a "day tripper," three a "certified AI explorer," six a "trailblazer," and 10 a "globetrotter." NLW's own verdict on the naming: "Is that the height of millennial cringe? Yes. Do I care? Not even a little." Link: https://aidailybrief.ai/e/2026-07-26#stamp-levels ### "Pack Your ID" — build a portable profile of yourself `[13:00]` A beginner "quick trip" project has you build a paste-ready global identity block of 150-300 words so any AI starts a chat already knowing who you are. Each lab ships with instructions, copyable prompts, an extension (e.g. teaching the AI how to push back on you), and a stretch goal. *For: Ops* Link: https://aidailybrief.ai/e/2026-07-26#pack-your-id ### Building your first app is the biggest unlock `[15:00]` NLW argues that vibe-coding your first application is one of the biggest unlocks available right now — more than almost anything else, it shifts you from seeing AI as an assistant that speeds up current work to something that unlocks entirely new capabilities. "Excursions" are projects that build something outlasting a single session. *For: Eng, Product* Link: https://aidailybrief.ai/e/2026-07-26#vibe-code-first-app ### "Lemonade Stand": build an AI-staffed micro business `[16:00]` This flagship "expedition" runs three sprints: mining 5-8 ideas from your history and skills down to a ranked top three; mapping one into a real micro business with an AI org chart and week-one task list; and validating demand with a thin demo before overbuilding. NLW says even happy employees gain from designing a full micro business. *For: Exec, Sales* Link: https://aidailybrief.ai/e/2026-07-26#lemonade-stand ### "The Loop": agentic loops for non-technical work `[18:00]` The Loop expedition helps you build an actual agentic loop in a useful area of work — harder and less intuitive than in software engineering. It front-loads background learning on what a loop is and isn't, costs, models, how loops fail, and how to run one in tools like Claude Code, Cursor, and Codex. *For: Ops, Eng* Link: https://aidailybrief.ai/e/2026-07-26#the-loop ### Everyone has a capability overhang `[19:00]` NLW's closing thesis: the 2026-defining idea is the capability overhang — the gap between what AI can do and what we're actually using it for. Almost no one is exempt, not even lab researchers or full-time AI content creators, because capabilities keep racing ahead faster than anyone can adopt them. The Summer Adventure is meant to help you close your own gap and have fun doing it. *For: Exec, Ops* Link: https://aidailybrief.ai/e/2026-07-26#capability-overhang *Today's sponsors: Robots and Pencils, Rackspace, Blitzy, Airtable (Hyperagent) — offers at https://aidailybrief.ai/sponsors* --- Transcript: https://aidailybrief.ai/e/2026-07-26/transcript.md Listen: https://pod.link/1680633614 · Ad-free: https://patreon.com/aidailybrief © 2026 The AI Daily Brief — Until next time, peace ✌