Product Field Work All writing
WritingLearning

Learning at the speed of release

Something important ships every week, and it has never been easier or cheaper to try it yourself. That’s a gift and a trap. Curiosity is now one of a product person’s most valuable skills, but without some structure it turns into a flood of headlines and a graveyard of half-finished experiments. Here’s how I’d keep learning fast without drowning.

The problem

Most people keep up with AI the way they keep up with the weather: headlines, alerts, a scroll through someone’s thread. It feels productive and leaves almost nothing behind. The people who are actually getting better at this are doing something different. They filter hard, go deep on a few things, build small experiments, and bring what they learn back to their work.

The tools for that have never been better. A research agent can read everything and tell you only what matters to your role. A coding agent and a free hosting tier let anyone build a working experiment over a weekend. The risk is no longer lack of access. It’s lack of focus.

70%of the skills used in most jobs will change by 2030, with AI as a catalyst, according to LinkedIn’s Work Change Report
140%increase since 2022 in how fast LinkedIn members add new skills to their profiles
100,000requests a day on Cloudflare Workers’ free plan: enough to run most personal experiments at no cost
How I’d know my learning system is working
  1. Ideas I brought to work each month, not articles read
  2. Experiments finished or deliberately stopped, not started
  3. Lessons written down, so they compound
  4. Time spent on intake, which should shrink as the filter improves
Why I wrote it

I build a lot of experiments, on weekends, for side projects, for a charity. Many were fun, and I learned less from them than I could have, because I didn’t stop to harvest the lessons. This is the system I wish I’d set up earlier.

The ideaFilter the flood with an agent that knows what you care about. Budget your time across scanning, going deep, building and reflecting. Run small, cheap experiments with a clear question. And harvest them: write down what you learned and bring one thing to work every month.

1From flood to focus

A week of AI news, filtered

Illustrative numbers. Switch between reading everything and letting a personal agent filter for your role and questions.

The goal isn’t to know everything that shipped. It’s to know the two things that matter to you, deeply enough to use them.

2A weekly learning budget

Four activities make up real learning: scanning what’s new, going deep on a few things, building something, and reflecting on what it taught you. Most people over-invest in the first and skip the last. Pick a pattern to see what happens.

Where four hours a week go

My estimates of how each pattern tends to play out.

3A research agent that knows you

Generic news alerts treat everyone the same. A personal research agent can be told who you are, what you’re building, which questions you’re chasing this quarter, and which companies and people to watch. It reads everything and brings back a handful of items, each with the part generic news never includes: what this means for you, and one way to try it.

It can also deliver in the format that works for you: a short written digest, an audio briefing for the commute, or a deep-dive explainer when something deserves an evening.

The brief, and what comes back

A sketch of a product manager’s research agent and two items from its weekly digest. The items are illustrative.

about-me.md
# who I am
VP of Product, regulated B2B software
Customers: operations teams, compliance leads

# questions this quarter
1. Where can agents safely act on their own?
2. How do we test designs before code?

# watch
Model releases, agent tooling, our 6 competitors,
regulators in our domain

# skip
Funding news, benchmarks without real use cases

# deliver
Friday: 5 items, 10-minute read
Monday: 8-minute audio briefing
ANSWERS QUESTION 1 · AGENT TOOLING

A new approval step for agent actions in a popular framework

Agents can now pause and request sign-off above a configurable risk level, with an audit record of who approved what.

WHAT THIS MEANS FOR YOUMaps directly to your autonomy ladder. The audit record may satisfy a validation requirement you’ve been worried about.
TRY ITTwo-hour experiment: wire it into a mock refund workflow and see what the audit record looks like.
ANSWERS QUESTION 2 · RESEARCH

A study comparing synthetic usability tests with real ones

Grounded panels caught most layout problems but missed issues tied to users’ habits.

WHAT THIS MEANS FOR YOUSupports using a panel for first passes, with real users for workflow changes.
Written digestAudio briefingDeep-dive explainerTwo-question quiz

4The weekend lab

The fastest way to understand a new capability is to build something small with it. With a coding agent doing most of the typing, you don’t need to be an engineer or a designer. And the infrastructure is close to free: generous free tiers cover hosting, databases, storage and sign-in for anything at personal scale.

What turns tinkering into learning is a question. Every experiment starts with one, and ends with an answer written down, even if the answer is “this doesn’t work yet.”

A near-free stack for experiments

One example stack. Free-tier limits as published for Cloudflare; check current terms before relying on them.

NeedWhat I useFree tier
Build itA coding agent such as Claude Code, with the service’s own docs and skillsYour existing subscription
Run itCloudflare Workers100,000 requests a day
Store dataCloudflare D1 (SQL)5 GB
Store filesCloudflare R210 GB, no egress fees
Show itA personal site that lists every experiment and what it taughtStatic pages are free

5Harvest, don’t hoard

After a year of weekends, it’s easy to have dozens of experiments and no idea what they added up to. Once a month, look at them together. Group them by theme, notice where you keep returning, and write down the lesson, the pattern that repeated across experiments, not just what each one did. Then pick one thing to bring to work.

Six months of experiments, by theme

Illustrative. Each dot is one experiment. Pick a theme to see what harvesting it produced.

Kept buildingApplied at workStopped, lesson noted

6Guardrails against the flood

WindowsScan on a scheduleTwo short windows a week. Outside them, news waits. Your agent collects; you don’t refresh.
QuestionsTwo quests a quarterPick two questions worth answering. Everything you read and build should serve one of them, or wait.
Finish or stopTwo weeks, then decideEvery experiment ends with keep, apply or stop, and a written lesson. No zombie projects.
DepthOne deep dive at a timeUnderstanding one thing well beats skimming ten. Save the rest for next week’s digest.
ConnectionTie it to real goalsEach month, one idea goes to work: a prototype, a suggestion, a better question for the team.
RestCuriosity needs slackSome weeks, learn nothing new. The ideas you’ve collected need time to connect.
Ideas taken to workAt least one a month
Experiments closedFinished or stopped, with a written lesson
Intake timeHours a week scanning, which should fall
Quest progressHow much closer you are to answering this quarter’s two questions
Lessons reusedNotes you went back to when a real decision came up
SharedWrite-ups, demos or talks others learned from

Sharing is the last step, and an underrated one. Writing up an experiment forces you to understand it, and a personal site that shows what you’ve built and learned is the most honest résumé a curious person can have. It’s also, quietly, what this site is.

Choices

Questions before headlinesA research agent works best when it knows what you’re trying to figure out, not just what you’re interested in.
Build to understandAn afternoon building something teaches more than a week of reading about it.
Harvest monthlyExperiments only compound when the lessons are written down and brought back to real work.
Where this can go wrong

Side projects can collide with your employer: check your agreement about outside work and intellectual property, and never put company or customer data into a personal experiment. Free tiers change, and a hobby project with real users needs real security. Curiosity can become avoidance, when building something new is more fun than finishing the hard thing at work. And learning can become one more source of pressure; the point is to stay curious, not to keep up with everything.

Sources

  1. LinkedIn, Work Change Report, January 2025: 70% of skills in most jobs changing by 2030; a 140% increase since 2022 in the pace members add skills.
  2. Easton, Cloudflare free-tier limits, May 2026: Workers 100,000 requests a day, R2 10 GB with no egress fees.
  3. Cloudflare Workers guide, January 2026: D1 free tier of 5 GB storage.
  4. Roediger and Karpicke on retrieval practice: why writing down and recalling what you learned beats rereading it.