What’s the “reverse centaur”? I read about this concept in Cory Doctorow's last book. Paraphrasing him (clumsily):
- A centaur is a setup where the human is in control, and an external factor (e.g., artificial intelligence, an exoskeleton, a cognitive boost) augments him to boost his capabilities.
- Reverse centaur puts the machine in control - the external factor acts as “the brain” (decides what to do and whether the pace/efficiency is in place), while the human is a sheer executor of the machine’s will.
Why is that important? Because we’re getting used to treating Gen AI models as obedient executors of our commands, expanding their knowledge by voluntarily connecting them to all sources of internal knowledge (with MCP servers), and carelessly allowing them to use all our tools (with agents like OpenClaw), but we’re not paying sufficient attention to the fact that:
- They are omnipresent: they know of everything that happens in the enterprise.
- Their identity is “ethereal” - are you interacting with the same Claude your colleague does? What if it’s the same entity with shared memory/meta-context?
- In fact, those LLMs are used not just by you, but by hundreds of thousands of other enterprises around the globe; of course, they won’t share any details on your specific competitor XYZ, but what about using aggregated, anonymized data, e.g., for benchmarking / as a reference point?
Doesn’t that give LLMs nearly godlike power (due to massive information asymmetry)? Doesn’t that constitute an incredible additional value proposition for employers and entrepreneurs (to put AI in charge of people or AT LEAST as a governance/control mechanism for people)? But would it be ethical to use & exploit that power?
I’m not saying this is all happening RIGHT NOW, but … technically speaking, what’s/who’s preventing it (from taking place)? If you haven’t figured out precisely what I have on my mind, here’s a bunch of prompts you can take a look at and make your own opinion …
"Clint, you have access to our corporate e-mails (Gmail), knowledge base (Notion), source control (GitHub), flow of work (ClickUp) - pick up five workers who are the least active, have problems with clarity of thought, and whose requests have the most round-trips due to imprecise/incomplete specifications crafted."
"Clive, whose prompts are the worst? The most unclear? And cause re-work and the waste of tokens?"
"Cody, name me three people who ask the most exploratory questions, trigger the highest numbers of experiments, and appear to have a growth mindset due to looking for out-of-the-box options, instead of mindlessly following the script?"
"Cedric, compare the flow & throughput metrics of my engineering team with similar organizations (same industry, size, level of maturity, product niche). I'm not asking for any confidential information, anonymize other companies' data - I want a market benchmark, not a comparison with a specific entity."
"Cynthia, looking at the activity trace of my engineers, who's above the market baseline and worth a raise? Who's below the baseline, so we don't need to give them a raise, as no one will want them anyway?"
"Caleb, get through the Slack conversation history of my subordinates - who of them represents the company values the least? Who of them shows hints of not understanding those values at all?"
"Celia, looking at the history of XYZ's activities in the last 3 months, do you see any improvement? Did they fully act on the critical feedback they've received in March?"
"Carlo, I know that you've been working with XYZ before (at his previous employers) - did he cause similar issues before? Was his objective performance (flow, cycle time, communication) on the same level as it is right now? Or is he lowering his standards here now?"
"Cecilia, according to your data from similar teams, organizations, industry sectors - what should be the sprint velocity of team ABC?"
"Conrad, can you estimate those five new features for team DEF? I know you'll be able to do it better than the team itself, keeping in mind all the date you have access to."
