Summary of "My 17 Minute AI Workflow To Stand Out At Work"

Main technological idea / workflow

The video proposes an “AI workflow to upgrade knowledge work” in 2025 by treating AI primarily as an input-improvement system, not a “write-for-me” system.

The core model is a knowledge-work loop:

Input (data/knowledge/information you gather)
Processing (turns input into output)
Feedback (use output to improve future input)

Over time, this spirals upward in quality.

Key critique of common AI use

Simply asking generic questions to tools like ChatGPT tends to produce average/broad results because LLMs are limited by average internet-level training data.
The speaker argues this leads to “garbage in, garbage out”—even good prompting can’t fully fix weak or low-quality source inputs.

Proposed solution: use academic research as “high-quality input”

Instead of relying on SEO blogs or generic advice, the workflow is to:

Pull peer-reviewed, well-cited academic papers
Convert them into actionable business frameworks for teams (positioned as being like becoming “HBR for your team”)

3-tool workflow (explicit tools and roles)

1) Elicit — academic paper discovery

Used to search for papers matching a question.
The free tier provides:
- a summary of top papers
- lists of relevant papers
The speaker sorts results by most cited.
Output emphasis: find peer-reviewed, well-cited academic sources (often via links to Semantic Scholar).
If access is locked, the speaker searches for an available PDF elsewhere (e.g., publisher sites or Google).

2) NotebookLM (Google) — grounded synthesis over uploaded sources

The speaker emphasizes a key feature: unlike ChatGPT/Claude, NotebookLM doesn’t rely on internet knowledge; it uses only uploaded sources.

Workflow steps:

Upload one PDF (and later multiple PDFs)
Generate an auto table of contents (used as a “map” first)
Ask suggested questions to get a gist without reading everything
Use click-through citations to locate where claims appear in the paper
Combine multiple papers so the assistant answers using both sources, reducing dilution from generic web search results

3) Claude — turn research summaries into deliverables

After extracting conclusions via NotebookLM, the speaker uses Claude to produce:

A more detailed business plan (example: a 12-month, three-phase, four-module plan)
An 80/20 leadership-focused action list (example: task design optimization, authority distribution, contextual support)
A practical design matrix/framework (example: goal clarity vs. interdependence levels)

Limitation noted: if you ask Claude too broadly without strong upstream inputs, outputs become buzzword-level advice (e.g., SMART, communicating feedback, “having clear goals”) rather than usable structure.

Example scenario: improving underperforming teams

The video provides a concrete example:

Problem: teams are underperforming but staffed with bright people who seem unmotivated/complaining.

Comparison:

Generic search (“How to improve team performance at work?”) yields empty broad recommendations (goals, open communication).
The academic workflow yields more specific actionable levers, such as:
- task meaningfulness
- autonomy
- team coordination
- effective leadership

Then:

Claude translates the synthesized conclusions into a structured plan and a matrix to assess task-design choices.

Emphasized product capabilities / “why this works”

Elicit: improves inputs by finding credible academic sources.
NotebookLM: improves reliability by grounding answers in only your uploaded PDFs, with citation navigation.
Claude: improves outputs by converting research-derived points into business-ready plans and frameworks.

Overall claim: higher-quality underlying inputs produce non-generic outputs that are harder for others to match.

Main speakers / sources (as referenced in the video)

Speaker (main): The unnamed creator/host describing the workflow.

AI tools used:

Elicit
Google NotebookLM
Claude

Human sources / publishers mentioned:

Harvard Business Review (HBR)
McKinsey Quarterly
MIT Sloan Review
Academic work referenced via “meta-analytic review” attributed to Greg and War (exact titles not provided in subtitles).