[Day 1] Using AI to Plan My Year of AI Experiments
[Day 1] Using AI to Plan My Year of AI Experiments
Vague goals fail. Here’s how I used NotebookLM and Claude to extract a behavioural scientist’s framework from a podcast and turn it into a concrete 365-day plan—complete with cue-based triggers, obstacle planning, and emergency reserves.
I've spent my career at the intersection of curation, human resources, industrial-organisational psychology, and technology. Now I'm asking a question that feels urgent: What does human work look like when AI becomes our constant collaborator?
🔎
This blog is my daily experiment log. Each post will document a tangible output I've created using AI tools—not to prove AI can do things for us, but to demonstrate how humans with curiosity, lateral thinking, and domain expertise can use AI to scale their impact.
The thesis I'm testing: AI is a powerful amplifier, but it still requires human creativity, judgment, and taste to produce genuinely useful work.
Today's Experiment: Planning Goals Using AI and Podcast Insights
For my first entry, I wanted to do something meta—use AI to help me plan how I'll conduct AI experiments throughout the year. I started with a podcast episode with Mel Robbins featuring Professor Katie Milkman on strategies for lasting change.
My Process
Step 1: Extract actionable insights from the podcast
I uploaded the podcast to NotebookLM and used this prompt:
"Provide a concise and accurate list of all the recommendations covered in this podcast interview about strategies for lasting change in a way that the reader can easily adopt and apply in real life."
Step 2: Dig deeper into what intrigued me
The phrase "make a concrete plan" stood out from the summary, so I asked a follow-up question:
"Make a Concrete Plan - what is the professor's specific advice to making a plan that will be effective?"
NotebookLM returned a structured framework with five key components:
Set a Concrete and Measurable Goal – Move from abstract desires to specific, measurable targets broken into daily actions
Use Cue-Based Planning – Link desired actions to specific triggers (time or event-based)
Engineer for Obstacles – Identify friction points before they arise and plan around them
Layer in Social Support and Fun – Add accountability partners and temptation bundling
Build in Emergency Reserves – Include flexibility days to prevent quitting after a single misstep
"How can I use this advice to create a plan to conduct experiments using AI that I can post daily to the blog?"
My 2026 Daily AI Experiment Plan
Here's the plan Claude helped me develop by applying Professor Milkman's framework:
The Goal (Concrete and Measurable)
Publish one blog post daily for 365 days, each documenting a specific, replicable AI experiment with a tangible output.
Success criteria:
Each post includes the problem, tools used, prompts applied, and the output
Posts are useful enough that readers could replicate the experiment
I maintain consistency across the full year
The Cue-Based System
Trigger
Action
7:00 AM coffee
Review experiment queue and select today's topic (I’m using Notion and will share more about this in a future post)
8:00 AM (blocked calendar)
Conduct the experiment using AI tools
12:00 PM (lunch hour)
Draft the blog post
6:00 PM
Edit, publish, and schedule promotion
The specificity matters. "I'll blog about AI" becomes "At noon, I sit at my desk and draft today's post documenting what I made that morning."
Engineering for Obstacles
Obstacle
Pre-planned solution
Runs out of experiment ideas
Maintain a running list of 30+ ideas, updated weekly
Low energy or sick days
Pre-write 5 "buffer" posts for emergencies
Travel
Schedule lighter "reflection" posts that don't require new experiments
Perfectionism slows me down
Set a 45-minute timer for drafting—publish "good enough"
Temptation Bundling
AI experiments happen only during my morning focus block with a cup of coffee
Post-publication reward: 15 minutes of guilt-free reading or podcast listening time
Weekly "fun experiment Fridays" where I try something purely creative with no practical requirement
Social Accountability
Share each post with a small accountability group
Monthly check-ins with a friend who's also exploring AI tools
Engage with blog comments and questions—the community becomes the accountability partner
Emergency Reserves
Instead of demanding 365 posts or nothing:
Target: 365 posts
Minimum: 300 posts (65 "emergency reserve" days)
Recovery rule: If I miss a day, I don't spiral—I just continue the next day
This prevents the "what-the-hell effect" where one missed day derails everything.
What I Learned Today
This process demonstrated something important: AI didn't give me the plan. I had to:
Choose the right source (the podcast)
Ask the right questions (zeroing in on "make a concrete plan")
Synthesize across tools (NotebookLM → Claude)
Apply it to my specific context (my background, constraints, goals)
The AI accelerated what would have taken hours of note-taking and research. But the judgment calls—what to focus on, what resonates with my situation, how to adapt generic advice—those remained mine.
Try It Yourself
Tool 1: NotebookLM (or similar audio/document analysis tool)
Prompt 1:
"Provide a concise and accurate list of all the recommendations covered in this podcast interview about strategies for lasting change in a way that the reader can easily adopt and apply in real life."
Prompt 2 (follow-up on what interests you):
"[Topic that intrigued you] - what is the professor's specific advice to [doing that thing] effectively?"
Tool 2: Claude (or similar conversational AI tool like ChatGPT or Gemini)
Prompt:
"How can I use this advice to create a plan to [your specific goal]?"
Then iterate until you have a plan you'll actually follow.
Every year since ChatGPT launched, I've found myself asking a completely different question about generative AI—not the same question with better answers, but an entirely new question that only became possible because the previous one had been resolved. In 2022, I wondered what AI actually knew (and learned the hard way about hallucinated salary data in job descriptions). By 2025, I was producing interactive training decks and Excel templates without writing a single line of code. Now, at the start of 2026, I'm asking how to build automated workflows that free me up to focus on what humans do best. These five questions trace the extraordinary pace of change in just five years—and reveal where the real opportunities lie for knowledge workers ready to ask the next question
I've tried using Claude, ChatGPT, and Gemini to create presentations and the results were always underwhelming. The content was decent, but the formatting? A manual nightmare. So when I tested Gamma.app with a genuinely complex use case (employee offboarding compliance across six Asia-Pacific countries), I wanted to see if a specialized tool could actually deliver what general-purpose AI assistants can't: polished, presentation-ready output in minutes, not hours. Here's exactly what happened, step by step, including the limitations they don't tell you about upfront.
What happens when you give three AI tools the same complex virtual training scheduling problem and one of them walks straight into a cultural landmine? I ran an experiment with Claude, ChatGPT, and Gemini: identical prompts asking each to recommend the best approach for scheduling a 100-person virtual training across six APAC countries during Lunar New Year season. One tool recommended a culturally tone-deaf date to save $2,500, and one framed the time zone constraint as a "Golden Window" that actually helped me think through the problem differently. This is what AI collaboration looks like in practice and why the human still needs to be in the loop.