The Curious Field Guide: When AI Gets Hands (and We Get Ads)
This week’s thread: AI isn’t just “answering” anymore, it’s acting, monetizing attention, and quietly changing what work feels like.
1) The Opening Scene
I keep coming back to a small moment that happens all the time now: I’m driving between meetings, talking out loud into my phone half brainstorming/half therapy, trying to turn word-vomit into something coherent I can actually write later.
Six months ago, I thought the story would be “AI saves time.” The slides all say it. The vendor decks all promise it. But in my lived experience, that hasn’t been the headline.
The headline is more honest (and more interesting): AI helps me see differently. It helps me process, reframe, remix, and catch patterns I would’ve missed. That’s not efficiency, it’s effectiveness. And if AI is growing hands (agents), growing a business model (ads), and growing our workload (intensification), then the real question isn’t “what tool should I use?” It’s:
Where do I want speed and where do I want friction? This is the slide I have been using a lot lately to levelset the learning and thinking around AI
Truth be told, as I am literally typing this newsletter out when I have to many things to do I am watching Claude Cowork organize my entire desktop with permission and my multitasking messy brain is blown away!
2) Signals (3 max)
Signal #1: Agents are growing hands
· What happened: The “agentic” shift is accelerating and tools like OpenClaw point toward AI that can do things, not just talk about them.
· Why it matters: The moment AI can click, install, post, book, and execute… we need permission models and guardrails that most of us haven’t built yet.
· My take: It feels like giving someone access to my house because they offered to help clean the kitchen. Helpful… but also: which doors did I just unlock?
· Try this: Before you delegate, write a “never allowed” list (passwords, payments, messaging, posting, contacts).
· Link:
Signal #2: ChatGPT is testing ads (Free + Go in the U.S.)
· What happened: OpenAI has flipped the switch and started testing ads in ChatGPT for Free and Go tiers. Ads are labeled “sponsored,” visually separated from answers, and OpenAI claims ads won’t influence responses.
· Why it matters: Ads change incentives. Even when answers stay independent, the product becomes an attention environment, not just a tool.
· My take: I’m not surprised at all and this is the most predictable plot twist in tech. The real literacy move now is: notice the incentives and design your habits accordingly.
· Try this: If you’re on Free/Go, open your ad settings once and treat it like checking privacy settings on a new app.
· Link: https://openai.com/index/testing-ads-in-chatgpt
Signal #3: The “AI saves time” story is cracking
· What happened: A new Harvard Business Review case study argues AI didn’t reduce work but rather it intensified it: faster pace, broader scope, and work spilling into more hours (often without being explicitly required).
· Why it matters: If productivity gains translate into higher expectations, AI becomes workload creep in a nicer interface.
· My take: This matches what I’ve been saying for a while: AI rarely saves me time. It helps me think better. That’s why I keep coming back to one framing question in PD: Are we solving an efficiency pain point or an effectiveness pain point?
· Try this: Run a Pain Point Audit: list one task that steals time (efficiency) and one area where students struggle (effectiveness).
· Link: https://hbr.org/2026/02/ai-doesnt-reduce-work-it-intensifies-it
3) My Stack This Week (Tool + Workflow)
The 3-Layer Stack (how my workflow changed)
Layer 1 — Operate (daily default): Google Gemini and Claude
My basic work engine now: quick drafting, day-to-day synthesis, and Deep Research when I need a solid overview.
Layer 2 — Build (big projects): Claude + Claude Code
My workshop for bigger builds: PD design, tool/workflow creation, structured writing, and “make this thing real.”
Layer 3 — Capture + Automate (infrastructure): ChatGPT + Copilot + NotebookLM
ChatGPT: voice capture, my meeting-notes workflow (Projects), automation experiments (agents/projects), and an archive of prior prompts/tools.
Copilot (Word/PowerPoint): finishing and formatting when I’m already inside Office.
NotebookLM: my research library for admin coursework, PDFs, and curated sources I want to return to.
What changed in the last 6–8 months (the real shift)
· I stopped “tool collecting” (Perplexity, Gamma, NapkinAI, random doodads) and started “workflow committing.”
· My default question changed from “What tool can do this?” to “What layer does this belong to?”
· ChatGPT moved from “front door” to “infrastructure”: capture, automation, and systems I already built.
One mini case study (how a deliverable moves now)
· Idea captured while driving → ChatGPT voice note to organize the mess
· Research + angles → Gemini / Claude Deep Research
· Build the real asset (PD outline, guide, tool) → Claude / Claude Code
· Finish + polish → Copilot in Word/PowerPoint
· Archive sources + keep it alive → NotebookLM
4) The Studio: One Repeatable Prompt (Steal this)
The Pain Point Audit Prompt (Efficiency vs Effectiveness)
Best used with: Gemini or Claude
Goal: Decide whether AI is being used to reclaim time or improve learning (very different implementations).
Prompt (copy/paste):
You are my AI implementation coach.
I want to identify two pain points:
A) EFFICIENCY pain point (time + repetition)
B) EFFECTIVENESS pain point (learning quality + student thinking)
Step 1: Ask me 5 questions to clarify my context (grade/role, constraints, tools allowed, time, audience).
Step 2: Help me define:
- 1 Efficiency Pain Point (specific task + frequency + time cost)
- 1 Effectiveness Pain Point (specific learner struggle + evidence + desired outcome)
Step 3: For each pain point, propose:
- A responsible AI use case (what AI does / what humans do)
- A guardrail list (privacy, accuracy checks, bias risks, what NOT to automate)
- A 20-minute pilot plan for next week
- A way to measure impact without extra busywork
Output as a one-page worksheet I can copy/paste into a doc.
My rule of thumb:
If the goal is “save time,” measure minutes reclaimed. If the goal is “improve learning,” measure clarity, engagement, transfer, or student agency and not output volume.
5) Classroom & Students (Opportunities + Projects)
Student Opportunity (shareable)
· Opportunity: America’s Youth AI Festival (Day of AI USA)
· Who it’s for: Students (varies by track)
· What they produce: AI-inspired art/performance and “AI for a better world” concepts
· Next step: Choose one track and build it into a 2–3 week studio project
· Link: https://www.dayofaiusa.org/festival
Classroom Project Prompt
Project: AI for a Better World, but with Constraints
Students pick a real local problem (school/community). They propose an AI-enabled approach and they must include:
· 1) what data it would require
· 2) who could be harmed by errors/bias
· 3) what human oversight is non-negotiable
· 4) a non-AI alternative (so AI is a choice, not a reflex)
· 5) one friction rule (where humans must slow down and decide)
6) Analog Anchor
Do one thing this week on purpose slowly: write the first messy draft by hand (or on a blank doc with no AI). Then bring AI in second as an editor, organizer, and challenger, not the author.
Closing Question
If AI isn’t saving us time, but it is changing how we think and how fast we work then what’s the one boundary you need to set this month to protect your attention?
Sources Mentioned This Week
OpenClaw / agentic shift (newsletter essay): OpenClaw
OpenAI announcement (ads): Testing ads in ChatGPT
HBR case study (work intensification): AI Doesn’t Reduce Work—It Intensifies It
Student opportunities: America’s Youth AI Festival (Day of AI USA)
If You Want the “Ads + Incentives” Angle
How OpenAI frames ads + who sees them: Testing ads in ChatGPT
Independent coverage (useful for context): The Verge — ChatGPT ads test
If You Want the “Agents Need Guardrails” Angle
Agent security + prompt injection defenses (browser/agent context): Anthropic — Prompt injection defenses
Practical risk taxonomy (industry standard): OWASP Top 10 for LLM Applications
If You Want the “Trustworthy / Responsible AI” Lens
The big framework (readable + widely cited): NIST AI Risk Management Framework (AI RMF 1.0)
Direct PDF (for teams who want the actual doc): NIST AI RMF 1.0 PDF
If You Want the “Students + Classrooms” Angle
Festival hub: America’s Youth AI Festival
Program overview (Responsible AI for America’s Youth): Day of AI USA overview
Background + launch news: Day of AI / MIT RAISE national movement
Optional “Context Links” (for readers who want more)
Work intensification discussion (secondary coverage, quick read): Gizmodo summary of the HBR finding