The Super Agency Practice Loop
Use AI on real work, add human judgment, and strengthen coordination muscles
- Difficulty
- Moderate
- Time to result
- ~ongoing to results
- Steps
- 5
- Confidence
- 83%
Hoffman's super-agency loop begins with a mindset shift: AI is amplification intelligence. The likely competitive divide is not simply people versus machines, but people who use AI well versus people who do not. Build the capability by applying AI to real tasks where you possess enough expertise to grade the answer. Inspect its weaknesses, iterate when useful, and add what the machine lacks: sharper judgment, specialized knowledge, creativity, and distinction. Practice expands beyond prompting into training, deploying, organizing, executing, and strategizing across agents. AI can also help with the transition it creates by suggesting suitable roles and teaching a worker how to perform them with AI support. Repetition builds the human-plus-AI operating muscle.
Origin
Hoffman developed the amplification framing in Impromptu and Super Agency and applies it through constant experimentation on investment analysis, market trends, and everyday tools.
Core principles
- 01AI amplifies human ability rather than merely replacing it
- 02The strongest competitor is often another human using AI
- 03Real work builds better judgment than novelty prompts
- 04Human expertise earns its premium through effective tool use
- 05Coordination matters as much as prompting
How to run it
- 1
Pick real work
Select a meaningful task in a domain where you can recognize a weak answer.
Pro tip Use investment analysis or market research rather than a novelty poem.
Watch out If you cannot evaluate the answer, practice may strengthen misplaced trust.
- 2
Generate and inspect
Use the tool, then identify what is basic, wrong, generic, or missing.
Pro tip Treat a poor result as information about where human judgment matters.
Watch out Do not accept fluency as evidence of correctness.
- 3
Add the human premium
Make the work sharper, more distinct, and better grounded in specialized experience.
Pro tip Name the contribution only you could make before editing.
Watch out Expertise that never engages with the tool may lose its practical premium.
- 4
Use AI on the transition
Ask which roles fit you with AI support and how to learn the human-plus-AI version of the work.
Pro tip Ask the system to identify where it performs badly and needs your help.
Watch out Do not restrict AI to producing outputs when it can also help navigate change.
- 5
Expand into coordination
Practice deploying, organizing, executing, and strategizing across specialized agents rather than relying on one prompt.
Pro tip Think like a manager of complementary collaborators.
Watch out More agents without clear intent can multiply noise rather than agency.
In the wild
A strong dock worker loses an old advantage when the forklift arrives. Unlike earlier machinery, an AI-like forklift could explain where to sit, how to direct it, what to watch for, and where it performs badly and needs human help. The disruptive tool also becomes the tutor for the new job.
→ The worker uses the technology to navigate the transition instead of treating displacement as the end of agency.
Despite extensive AI experience, Hoffman continually tests tools on investment analysis and market trends. When an answer is weak, he sometimes iterates and sometimes uses the system elsewhere. He also enables voice agents on his phone to discuss objects through the camera. The repetition teaches both capability and limitation.
→ Specialized knowledge compounds with tool fluency rather than standing apart from it.
Common mistakes
Practicing only on novelty tasks
Low-consequence prompts reveal little about how the tool performs on work that matters.
Defending expertise from the tool
Static expertise can become less valuable than expertise multiplied by effective AI use.
Stopping at prompting
The larger advantage comes from organizing, deploying, and coordinating capabilities around a clear goal.
Is it for you?
Best for
Knowledge workers and entrepreneurs who can judge output quality in a domain they understand.
Not ideal for
High-stakes tasks where the user lacks enough expertise to detect incorrect or unsafe output.
From the transcript
“AI means amplification intelligence”
“Can the AI help me with a transition?”
“We've moved from the computers being the bicycle of mind to AI being the automobiles of the mind.”
From the episode
Reid Hoffman: Superagency, How AI Will Help Humans Dominate the Future
Reid Hoffman