Use AI As Leverage On Your PM Time
Assign the model a role, feed it more context than you could read, and iterate the prompt until it works.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 4
- Confidence
- 86%
Yehoshua's practical playbook for PMs to get compounding productivity from AI: always be using the new products, prime the model with a role, feed it corpora too large to read manually, and iterate the prompt rather than expecting it to work out of the box. The payoff is doing high-value synthesis work that previously would simply never have been done.
Origin
Tamar Yehoshua's own and her Glean team's usage patterns, with the Discord-transcript example from a forward-thinking PM she knows.
Core principles
- 01Good PMs always use new products; AI is no different from mobile—use ChatGPT, Glean, Claude, try them all.
- 02The opening of the prompt matters: giving the model a role ('you are a PM at Glean') is far more powerful than 'summarize this.'
- 03Feed it context too large to read manually—entire channels, transcripts, news dumps—to unlock work you'd otherwise skip.
- 04It won't work out of the box; you must tweak the prompt and iterate with patience.
- 05AI is leverage on your time, so 'I'm too busy' is exactly backwards.
- 06AI is not good at being creative—use it for the route/synthesis work and keep the strategic, creative leap for yourself.
How to run it
- 1
Prime with a role
Begin the prompt by assigning the model a perspective—'you are a product manager at Glean, from that perspective give me this summary'—instead of a bare 'summarize this.'
Pro tip The role framing is the single change most people miss and it materially improves output.
- 2
Feed it more context than you could read
Dump large corpora the model can now hold—an entire Discord channel, all your sales-call transcripts, the day's industry news—and ask targeted questions of it.
Pro tip Great questions to ask a large channel/transcript: what's the sentiment, the most-requested feature, and what people are unhappy with.
- 3
Iterate the prompt until it's right
Expect the first version to fail (e.g. summaries that can't tell a customer request from a salesperson's suggestion) and tweak until it distinguishes what you need. Have patience and keep trying.
Watch out It won't work out of the box—teams that try once, fail, and revert to the old way get left behind.
- 4
Reserve the creative leap for yourself
Let AI surface what customers are asking for and today's problems, but do the creative, strategic synthesis—how to actually solve it—yourself, since that's what differentiates a great PM.
Watch out PMs doing mostly route/execution work risk being automated; the strategic, creative ones thrive.
In the wild
A forward-thinking PM took the entire (huge) transcript of his product's Discord channel, fed it into Gemini after the context window expanded, and asked for the sentiment, the most-requested feature, and what people were unhappy with. He called it 'a gold mine'—work that, done manually, simply would never have happened.
→ Instant synthesis of a corpus no one would have read, turning an ignored channel into prioritized product insight.
Yehoshua's team built a Glean app that reads all their Gong sales-call transcripts, puts them in a spreadsheet in a set format (who the AE is, etc.), and summarizes the top requested features. The first prompt was too weak—it surfaced features salespeople recommended and couldn't tell those from actual customer requests—so they tweaked it until it worked.
→ A reliable, time-saving pipeline of customer-requested features, achieved only after prompt iteration.
Yehoshua wrote a Glean prompt that takes a feature name and pulls the launch calendar date, open Jira/geer tickets, relevant Slack conversations, and beta-customer feedback into one view, giving her a confidence level on the launch—so she doesn't have to read all the channels herself.
→ One prompt replaced manual trawling across launch calendar, tickets, and channels to assess ship confidence.
Is it for you?
Best for
Product managers who feel too busy to adopt AI and want concrete, high-leverage workflows to start with
Not ideal for
Tasks requiring the genuinely creative strategic leap, which the models don't yet do well
From the transcript
“use the products like this is what good PM should do period always be using new products”
“the power of just giving it a role like you are a product manager at glean from that perspective give me this summary versus just…”
“he took the transcript from the Discord Channel which was huge right and he fed it into Gemini the entire Channel”
“it's not going to work out of the box but then we got it to the point where it worked”
“the one thing it's not good at is being creative”
From the episode
Lessons in product leadership and AI strategy from Glean, Google, Amazon, and Slack
Tamar Yehoshua (Product at Glean, ex-Google and Slack)