DRAG Framework
ai-literacydelegationproductivitythemitmonk-framework
DRAG Framework
Sandeep Swadia's 4-category delegation rubric for what to outsource to AI. From Dangerously Smart with AI (theMITmonk). Companion to the capped-payoff vs uncapped-payoff curves (intelligent laziness): DRAG is what zone 1 (capped) work looks like in practice.
The four categories
| Letter | Category | Why AI is good at it |
|---|---|---|
| D | Drafting | Solves the blank-page problem. "It's hardest to get from zero to one sometimes." First draft will be "crappy and atrocious, but that's fine. Now you have a starting point." |
| R | Research | Deep-research features (ChatGPT / Gemini / Claude) fire hundreds of secondary queries and consolidate results. "It's like you just hired a consultant for a week-long research project, but instead, you get there in 10 minutes." |
| A | Analysis | First pass at summarizing, reasoning, finding patterns — especially in unstructured data |
| G | Grunt work | Reformatting, translating, tabulating, cleaning data. The boring manual work. |
For the drafting case, Sandeep pairs DRAG with AIM Protocol (Actor / Input / Mission) as the prompting micro-pattern.
The decision rule
"Apply it only when you are in your zone one, that first curve. If it requires human interaction or judgment or intuition or decision-making or taste, that's curve two. That you've got to do it yourself."
Sandeep's estimate: "70 or 80% of my repetitive tasks tend to be in zone one."
How it relates to existing concepts in this vault
- Bounded vs Unbounded Tasks (Praveen Akkiraju) — the enterprise-agent counterpart. Praveen's bounded/unbounded axis is the enterprise-architecture frame for the same delegation decision. DRAG is the personal-productivity version. Useful because it gives the senior-IT-leader user a vocabulary for both sides of the same call (enterprise agent strategy and personal AI usage).
- Code Is Free — what makes DRAG inevitable. If implementation (drafting/grunt) is the cheap part, the human's value moves to zone 2 (judgment/taste). DRAG is the personal-time-allocation consequence of that pricing shift.
- Intelligent Gym — the inverse. DRAG outsources information work to AI; the Intelligent Gym keeps transformation work in-house and adds friction. The two together define when AI is a hire vs a spotter.
Where DRAG breaks down
Implicit risks Sandeep doesn't quite name:
- Drafting → Hallucination Laundering if the draft isn't verified before submission. DRAG's "drafting" step is safest when the human still does the heavy editing pass.
- Research → fabricated citations, especially with deep-research features that confidently cite papers that don't exist. The video's earlier "is this verified?" follow-up move is the necessary patch.
- Analysis → over-trust in synthesized patterns that the human can't audit. "AI is going to find patterns that we humans aren't going to be able to" is the upside; it's also the audit blind spot.
Applied in this vault
- DRAG for AI Upskilling at Manila IT Site — query (2026-05-30): the 4-phase rollout playbook for using DRAG as the spine of an enterprise IT team's AI-upskilling program. The practical-application page when the user asks "how do I use DRAG in my org?"
Sources
- Dangerously Smart with AI (theMITmonk) (canonical)