Video Summary - AI era skills: Why cultivating agency matters more than job titles

Business & product-building takeaways (AI era)

Max argues that traditional skill barriers matter less than agency—the ability to change outcomes and shape the work around you.
He dislikes the framing that “everyone’s the same,” but supports blurring in practice (designers and PMs contributing directly to prototypes and, increasingly, code).

Agency-building signals at Notion (examples)

Employees “drive Notion like it’s stolen” (even though they aren’t founders) by making concrete changes and reshaping how work gets done—not merely asking for permission.
Examples:
- Brian Leven: high-agency behaviors include recruiting heavily for what the org needs and blurring engineering/design boundaries.
- Eric Lou: shifted from long PRDs/strategy docs toward building prototypes first (e.g., spending more time in Figma to show/validate direction rather than only document).

“Prototyping over PRDs” / demos not memos
- Equivalent idea: send people something to react to (a build/demo) instead of a document.
- Strategy: use early AI-assisted prototypes to increase the number of testable hypotheses.
“Taste” as a product capability
- Taste = running a “virtual machine” to predict how a defined audience (in-group) will react.
- Built through reps: treat product judgment like training a model (feedback loop).
“Tiny core” / superpower principle
- Great products win because they have one tiny core that is exceptionally good (and markets amplify it).
- Product failure mode: “death spiral” of adding features—“one more thing” never makes it great.
Jobs-to-be-Done (JTBD) used pragmatically
- Use JTBD as a reminder to ask: What are users hiring this product for?
- In large orgs, teams stop “thinking like users” and start reviewing from an internal employee perspective.

Because janky demos/prototypes are cheap and fast with AI tooling, teams can front-load thinking:
- Write less PRD-heavy documentation for early exploration
- Produce early “version 0.8/0.1” experiences quickly
Framing: the last 10% is still ~90% (hardest part: correctness, quality, reliability).

Teams can afford to explore multiple directions.
Example behavior: “send off 10 agents to explore 10 different things,” then converge.

He’s not primarily optimizing for designers shipping production code.
Instead, coding matters because it forces mastery of the medium (especially agent loops), improving the quality of design decisions that engineering must later implement.

Despite faster shipping, he says software quality (reliability) hasn’t improved enough over the last year—more software exists, but finding truly reliable software is still hard.
He emphasizes:
- Incremental correctness (iterate toward correctness; consolidate toward stability)
- Consolidation across “evolutionary branches” of automation primitives (avoid fragmentation)
Customer feedback loop example:
- Notion has “six automation primitives” (including agents/automation types) and faces ongoing work consolidating them into a clearer core to reduce confusion and duplication.

Malleable software = software designed to work closer to users’ interests than the corporation’s interests.
Business implication: users should be able to rearrange/modify workflows (ownership of “computing life”).

Physical-world constraint vs user control:
- If you couldn’t rearrange your living room, you’d never accept it—software currently behaves similarly (apps tightly glue UI + data + behavior).

He argues the “SaaS apocalypse” (build everything from scratch; abandon SaaS tools like Notion/Slack/Salesforce) is overstated.
Reasons:
- Users don’t want full-stack maintenance (like not “hunting” for a steak)
- SaaS’s value is maintenance + specialists
Expected direction:
- Tools become more general again (toward “operating system / spreadsheet / word processor” eras)
- Still delivered “as a service”
- Specialized layers remain (security, compliance, enterprise readiness)

Token spend
- Notion policy described as effectively “unlimited” for now (not optimizing for spend).
- He hints many companies will start ROI pressure in 6–12 months (not a hard number, but an expectation).
- He avoids exact per-user token figures; estimates “thousands,” possibly “tens of thousands” for individuals, but doesn’t claim precision.
Feature iteration
- He references “shots on goal” as an internal operating metric (more experiments/testable outputs).
- He rejects “feature count” as a useful metric by itself (analogous to lines of code / token consumed).

Replace early PRDs with prototypes/demos
- Use AI to quickly create “reactable” versions; iterate earlier.
Engineer “the consolidation step”
- When multiple automation primitives/agents proliferate, plan for consolidation into a stable core experience.
Optimize for “obviously good”
- His standard: make something so good it doesn’t require a persuasive pitch—users immediately recognize value.
Build taste with frequent reps + feedback
- Create/tinker/shipping loops; surround yourself with high-quality references (and environments that raise the craft bar).

Presenter / guest: Max Schoening (aka Max Shying / Max Shing in subtitles), Head of Product at Notion; former Product Manager at Google, design/product leader at Heroku, design/engineering at GitHub.
Host: Notion podcast host (subtitle does not provide a clear name; the host repeatedly references the podcast and asks questions).