You’re not ambitious enough with Claude

The biggest productivity & capacity unlock in my daily work happened when I went beyond automating busywork to asking Claude to do more high leverage work.

For months, I was using Claude to clear the dopamine backlog, that queue of little tasks that I respond to too eagerly because completing gave a dopamine hit, an illusion of progress.

These automations freed up a lot of my time. With that new found time (and Fable), I started pushing it for higher leverage work - hard tasks that I didn’t trust Claude with before. This made the biggest leap in my productivity, capacity, and sense of flow.

Using Claude for all 3 levels of product work

@shreyas has a lovely concept of three levels of product work: impact, execution, and optics. Many of us use Claude to automate busywork (in optics & execution) and leave the impact work to ourselves. We tell ourselves the story that “judgment” is the defensible human thing. But when given the right context, Claude can do the hard work better and faster than you. Lowering the activation energy of starting is a major unlock in itself.

@NotionHQ surveyed 6,000 professionals across 10 markets on how orgs are using AI at work. At an individual level, if your org provides Claude/ChatGPT and connectors/MCP access, you can push your personal system further to the right and use Claude for all 3 levels of product work.

Notion@NotionHQ·Jun 25Replying to @NotionHQOnly 2% of organizations are operating at the most advanced level.

88% are still using AI mainly as a thought partner or assistant.

AI at most companies looks like using a chatbot - for drafts, summaries, even brainstorms. Which means there’s a lot of room to grow.295911K

As a platform PM, I need to understand and work with the entire product org’s roadmap (thousands of PMs and developers). Here’s how I use Claude at each level of product work that’s allowed me to scale up my scope and scale down my stress.

For Optics work: use Claude as autopilot to replace production.

Optics work is about making your team’s progress visible to the right people in the right forums. I’ve automated the majority of the production work and simply review/edit the output to use for different forums.

I have a scheduled task on Cowork that runs twice a week, it pulls context from key sources (Slack channels, calendar, meeting notes, git) and writes the latest updates for each work stream into a Google Sheet template. This G-sheet status board is visible for the teams to get their questions answered without asking me. I also have a skill that packages these updates at different levels for the right audience and forums.

Optics work is the layer to automate ruthlessly because it takes care of the dopamine backlog so you can stay in flow. I wouldn’t be surprised if these flows are productized soon in Claude itself.

Scheduled task for optics work

Scheduled task for optics work

Sample instruction template for scheduled task - iterate your own

markdown You are updating the [your status sheet] for [name]. Your job is to gather the latest updates from Slack channels, a Google Sheet tracker, and the PM journal, then write a JSON update file to a Google Drive inbox folder (which an Apps Script time trigger picks up and applies to the sheet), and send [name] a compact Slack DM.

STEP 1: Read current sheet state

[use google connector here]

STEP 2: Gather signals (last 3 days)

2a. Slack channels

Read these channels using slack_read_channel (oldest = 3 days ago unix timestamp): [insert your slack channels]

2b. PM Journal

Read the last 3 days of journal entries from the [folder]. These are local files at: [insert filepath] Read the 3 most recent .md files (by filename date). Journal entries contain to-dos, decisions, meeting notes, and context that may not appear in Slack. Treat journal entries as a first-party source alongside Slack messages.

STEP 3: Determine changes

For each track and detail row, identify ONLY items with verifiable changes:

• Status changed? (e.g., “◉ In Dev” → “Complete”)
• New items shipped?
• Next milestone updated?
• New or resolved risks?
• Notes worth updating?

Be factual. Only report changes backed by [your sources]

STEP 4: Push updates to the sheet via Drive inbox

Create a JSON file in the Drive inbox folder. The Apps Script time trigger (runs hourly) will pick it up and apply the changes to the sheet.

[work with claude on JSON payload]

STEP 5: DM Christine

Send a Slack DM to [yourslackID] with a compact summary. Format:

At-a-Glance sheet updated — [date]

[N] changes pushed: • [Track]: [one-line summary of what changed] • [Track]: [one-line summary]

Heads up: [1-2 sentences on anything notable that doesn’t map to a sheet row — new risks, scope changes, staffing, cross-team signals]

If nothing changed across all tracks, send: “At-a-Glance sheet checked — [date]. No changes detected across any tracks.”

GUIDELINES

• Be factual. No inferences or speculation.
• Use the sheet’s own terminology: “◉ In Dev”, “○ Planned”, “Complete”, “On Track”, “At Risk”, “In design”
• If a Slack channel has no relevant messages, skip it silently.
• Keep the DM under 500 characters. Christine wants glanceable, not comprehensive.

For Execution work: use Claude as copilot to accelerate execution.

Every Monday, I run the context-dump skill (click to install) which reads my past week across Slack, calendar, notes, and repo activity, gives me an analysis of my efficiency and patterns to watch, recommends what I should prioritize, and offers specific work to help me get started on.

mocked sample of context-dump skill’s output

mocked sample of context-dump skill’s output

It reads less like a summary and more like a coach who notices my personal patterns and actually makes helpful observations and recommendations. Here’s one from last week:

“Leverage dipped below the ~70% target this week, not because overhead crept up (it’s at the line), but because alignment/ceremony load (~35–40% of calendar) crowded out heads-down leverage work. The risk is that too much of it happened in meetings rather than in written, shareable artifacts.”

I also run a planning skill that checks Jira status and the roadmap and shows me how the team is tracking, what to line up next. I’d then ask Claude to make those Jira updates or turn PRDs into tickets. I know this is awesomely automated in Linear but for those still using Jira - there’s no reason to manually create tickets now.

For staying close to the customer, I run a scheduled task monthly that analyses our customer support channel to give me the top themes, the trends, and recommends the top requests or problems I can prioritize.

Get more ambitious with the asks. Don’t treat Claude as just for distilling a summary, ask for judgment.

For Impact work: use Claude as sparring partner to get the hardest work started

This level is the high leverage work that is the hardest to start - eg. thinking through a new product bet, preparing narrative for a high stakes presentation, testing my strategy against business unit strategies, etc.

I no longer wait for the perfect condition (like a full 2 hour focus block) to start. Even with a 10 minute break, I can quickly voice dump in a Claude project (that’s loaded with my team’s and product’s context and a writing style guide) and spar on Fable high or max to get a shitty first draft. This used to take days.

It’s worthwhile to “onboard” Claude and set it up with the right context. I started with existing product decks, roadmap, our code base summary, then voice dump our goals & metrics, my stakeholders (their roles & what they care about), what I want to learn & grow in, etc. Then I’d work with this context folder in either Cowork project or in Claude Code for my tasks so it always gets the right context and also automatically updates that context as I work.

example of context folder

example of context folder

Judgment is mostly a function of context. A PM’s judgment is pattern-matching over the context they have. As a platform PM executing on 5 to 10 work streams at any given time each with a different team with different timelines and scope, I’m supposed to be the context router. Claude can hold way more context at once, so can certainly offer valuable judgement. (Obviously, it’d miss the context that’s not written down, like team dynamics and incentives, and it doesn’t have accountability for its ideas.)

Using HTML was another breakthrough in helping me understand Claude’s output better and faster than a wall of markdown text. It also gives me ideas on how to visualize my POVs during brainstorming. Simply ask Claude to “generate the output in a HTML artifact”. However, I never share html artifacts. It’s not collaborative, it’s another url for people to keep track of, and subconsciously it feels less personal as it’s obviously “AI-generated”. But structuring an idea visually takes a huge mental load off already. It gives me a starting point to then iterate in Claude Code via Figma connector to build the visual for the doc I actually share.

Fable is noticeably better than previous models for impact work. I especially enjoy its calmness, it sounds more concise. It feels more like a conversation with a smart colleague than reading walls of text with unnecessary dispositions. This makes a big difference in keeping a train of thought going.

Personal growth

Growing technical fluency

As someone who doesn’t have a CS degree, Claude Code is incredible in helping me understand technical concepts. I’d ask questions about a code base to understand its structure, generate an overview.md with architecture and feature list and reference it to debate on PRDs and roadmap. The immediate feedback on feasibility and technical approach helps me make decisions and plan way faster. I’ve pushed small PRs for the first time and had a taste for the real developer’s end to end workflow. I’ve learned a ton from just reading CC’s plans, seeing how it thinks through a problem. It gives me a great starting point to have a meaningful discussion with the engineers.

Seeing patterns & blindspots

The bonus benefit of the weekly context-dump is using the cumulative history to get honest self assessment for personal growth. It has the perfect rolling history of what I worked on, the challenges I encountered, accomplishments I made, and even interpersonal dynamics built up from meeting notes. Every few months, I’d ask Claude to review my rolling context-dump file and give me an assessment on new skills I’ve developed, blindspots I should pay attention to, and recommend me specific and actionable growth areas to work on to get to the next level (you’d have defined what your next level is in your goals.md in this project).

It’s like having a personal career coach to guide you. It’s also handy for writing your monthly and year-end reviews.

The setup (the least important part)

Beware of the second trap coming out of this - automation as a dopamine backlog. Don’t let the “perfect setup” become another distraction from just getting started. With that disclaimer, here’s my basic setup.

• Obsidian: a folder of daily markdown notes where I keep daily notes including meeting summaries (from your AI meeting summary of choice)

• Claude Code in VSCode: a workspace with my team’s repos, Q&A to understand the codebases and spar on technical ideas; skills to understand our latest releases and write prds; Figma connector to make visualizations for docs; scheduled tasks for automations through internal MCP connectors

• Claude desktop: Cowork for scheduled tasks; chat projects for brainstorming and writing, personal writing guide md file in the project folder

• Connectors: Slack, Google Drive, Outlook, Jira, GitHub, HeyMarvin

• PM folders: store evergreen markdowns that are important context, eg. context folder (with meeting notes), writing style guide, company annual goal, templates/frameworks etc.

Getting over yourself

Given the progress of models over the past few months and how impressive Fable is, it’s inevitable that Claude is going to be increasingly capable of harder tasks and do them better than us. I’ve come to admit that Claude with enough context can already outperform my thinking in many ways. The best you can do now is to keep pushing the model with higher expectations, ask it for unknown unknowns, and get the most out of it for all 3 levels of product work. Test it yourself: pick the highest leverage task you’ve been putting off, give Claude the context it would need, your current thinking, and ask what you’re missing.

If you want to get these directly in your inbox and see past writings, I also publish on substack.