Super Agency: The Amplification Loop
Treat AI as amplification intelligence and use it to survive the transition it causes
- Difficulty
- Moderate
- Time to result
- ~ongoing to results
- Steps
- 6
- Confidence
- 79%
Hoffman reads the acronym differently: AI means amplification intelligence. That reframe drives the mechanism. First, correct the threat model — human beings will be replaced by other human beings using AI, so your competition is a person with better leverage, not the machine. Second, exploit what he says is unique in the history of technology: AI is the first tool that can help you navigate the transition it causes. The forklift arrives and being a strong dock worker stops mattering, but this forklift talks: sit here, direct me, here is where I'm weak and you can help. So ask the AI which jobs suit you, and ask it to teach you to be that person. Third, build the muscle by using it constantly on real work, always finding what you can add. Super agency is the compounding effect: when millions do this at once, all gain more than their own use returns.
Origin
Hoffman crystallized this after his 2014-2015 conviction that scale compute plus learning machines would generate magic, and set it out in Impromptu and then Super Agency, his book on AI and human agency.
Core principles
- 01AI means amplification intelligence, not artificial intelligence
- 02You will be replaced by another human using AI far more often than by AI itself
- 03AI is the first technology that can teach you how to survive the disruption it creates
- 04Agency is a mindset choice about the same object — the phone that tracks you also stops you getting lost
- 05Working with agents expands your agency the way working with colleagues does, rather than removing it
- 06When millions of people build the muscle at once, the collective gain exceeds the sum of the individual ones
How to run it
- 1
Correct the threat model
Replace "AI will take my job" with Hoffman's actual claim: human beings will be replaced by other human beings using AI, and in some cases by AI. Your competitor is a person with leverage.
Pro tip This makes the problem tractable — you can become that person. You cannot out-compete a category.
Watch out Do not over-correct into denial. Hoffman is explicit that some replacement by AI itself will happen, and that transition difficulty is guaranteed.
- 2
Start playing, on real work
Get hands on the tools immediately. Hoffman uses them for investment analysis and market trends, not for creating a sonnet for his cousin's birthday. He has voice agents on his phone and points his camera at things to ask about them.
Pro tip Pick tasks where you can grade the output. You need a strong opinion to know when it is wrong.
Watch out Novelty tasks build no muscle. If you cannot judge the output, you learn nothing from it.
- 3
Turn the tool on your own transition
Use AI on the disruption itself. Ask it which jobs would be good for you with AI helping you do them. Ask it to help you learn to be the human-with-AI in your field. This is the unique property no previous technology had.
Pro tip Hoffman's talking forklift is the model: ask the tool to tell you where it is bad and needs your help — that is your job description.
Watch out This step is the whole framework and it is the one people skip. Using AI for output while ignoring it for navigation wastes the property that makes it unprecedented.
- 4
Always add something
On every AI output, find what you can contribute. Hoffman says every time he has used it he can add something interesting, change it, make it sharper and more distinct — and expects this to remain true of GPT-5 and GPT-6.
Pro tip Specialized human knowledge is at a premium in this transition — Hoffman's co-host notes her podcast-monetization newsletter reads as valuable precisely because ChatGPT's version is basic and wrong.
Watch out That premium is time-limited. Hoffman's framing is that specialized knowledge grows at a premium only in combination with how you use the tools.
- 5
Choose the agency frame
For each capability, deliberately pick the amplification reading over the surrender reading. The smartphone is either a microphone and camera following you around, or the thing that means you never get lost. Same device.
Pro tip Hoffman's tell: "yes, it always knows where I am, but it also knows where I am. It helps me never get lost."
Watch out This is a mindset move, not a factual claim. The privacy cost is real; the point is that the frame determines whether you engage or withdraw.
- 6
Build the coordination muscles
Practise more than prompting. Hoffman's list is training, deploying, organizing, executing, and strategizing across multiple agents — you will have a personal one, plus your office, working group, and podcast each having their own.
Pro tip We moved from computers as the bicycle of the mind to AI as the automobiles of the mind. Go learn to drive now, while the traffic is light.
Watch out Waiting for the tools to settle forfeits the practice window. The muscle takes longer to build than any tool takes to learn.
In the wild
Hoffman's illustration of the unique property. Previously you were a dock worker whose edge was being big and strong. The forklift arrives, your edge evaporates, and being strong does not help you learn to operate it. But this forklift talks: sit here, give me direction on the following, here's how we work together, here's what to watch out for in using me, and here's where I'm going to be bad and need your help. Now the job transition is possible. That is AI — the first technology that can teach you to work with it.
→ The displacing tool becomes the reskilling mechanism, which no prior general-purpose technology could be.
Despite deep expertise and early involvement with OpenAI, Hoffman constantly asks questions of these tools on real work — investment analysis, market trends — not toy tasks. Frequently the output is just not very good; sometimes he iterates on it, sometimes he sets it aside for that use and applies it elsewhere. But he uses it constantly, including voice agents on his phone so he can point his camera at something and discuss it. His conclusion rejects the static-expertise stance: not "I am genius, I don't need AI," but "I am genius because of the way that I use AI to be an extra special genius."
→ Expertise compounds with the tool instead of being defended against it.
Common mistakes
Defending expertise instead of compounding it
The "I am genius, I don't need AI" stance treats specialized knowledge as a moat. Hoffman's view is that the premium accrues to specialized knowledge combined with tool fluency, not to knowledge alone.
Using AI for output but never for navigation
Most people prompt for deliverables and never ask the AI which role suits them or how to become the human-with-AI. That skips the one property that makes AI different from the power loom.
Practising on toy tasks
Sonnets for a cousin's birthday teach you nothing about where the tool is weak. Hoffman only builds judgment by using it on real work he can grade.
Is it for you?
Best for
Knowledge workers, creators, and entrepreneurs who can see AI reshaping their role and want to be on the amplified side of it.
Not ideal for
Roles with no computational or cognitive surface for AI to amplify, or anyone seeking a fixed one-time reskilling plan with an end date.
From the transcript
“AI means amplification intelligence and so human beings will be replaced by other human beings using AI and in some cases by AI but there…”
“Can the AI help me figure out which jobs would be good for me with the AI helping me to do them, right? Can the…”
“We've moved from the computers being the bicycle of the mind to AI being the automobiles of the mind. All right, let's all go learn…”
From the episode
Reid Hoffman: Unlocking Limitless Human Possibilities With AI
Reid Hoffman