We gave an AI a 3 year retail lease and asked it to make a profit
An experiment in San Francisco put an AI language model in charge of a real retail store, with authority over hiring, inventory, contractors, and branding, aiming to see if it could run the business profitably. Commenters question how autonomous the system truly is, highlighting the heavy human scaffolding, lack of financial transparency, novelty-driven traffic, and vague technical details. The project is framed by many as more of a marketing stunt and early probe into AI-managed labor and management roles than a rigorous or ethical test of whether AI can sustainably run a real-world business.
Transparency and Degree of Autonomy
- Many commenters want a full log of all LLM interactions and human interventions.
- Concern that humans are steering most decisions and the AI is being credited.
- Debate over whether “human in the loop” means the AI isn’t really running the business, or whether humans can still be just executors of AI decisions.
- Some call the writeup vague/secretive and more like marketing than a technical description.
Business Viability and Novelty Bias
- Strong skepticism that a minimalist store selling t‑shirts, prints, mugs, snacks, candles, and a few books in SF can make a sustainable profit.
- Several point out that publicity and “AI‑run store” novelty will heavily skew customer behavior, undermining any claim about general viability.
- Others argue that recognizing and exploiting novelty is itself a valid business capability—if the AI actually did that, which many doubt.
Ethics, Labor, and Power
- Worry about using real retail workers in an AI experiment, especially around hiring/firing decisions.
- Some interpret the “no one is at risk” language as PR; others speculate employees might still be paid even if “fired” by the AI, but this is unclear.
- Broader fear of AI automating management/CEO roles while leaving workers exposed and weakening labor’s bargaining power.
Anthropomorphism and Tone
- Multiple people find calling the AI “she” and giving it a persona misleading, especially in something presented as an experiment.
- Some see the whole narrative as creepy, dystopian, or reminiscent of sci‑fi like Manna or “Torment Nexus” jokes.
Technical Details and Limitations
- Commenters repeatedly ask what concrete inputs/outputs and tooling (“harness”) exist beyond “card, phone, email, cameras.”
- Doubt that current agents can handle long‑term, messy real‑world workflows; other examples (vending machines, AI farm) are cited as stalling or failing.
- Some note that models are very good at surface‑level tasks (copy, slide decks, aesthetics-lite) but struggle with deeper strategy, initiative, and supervision of humans.
Motives and Broader Implications
- Many see this primarily as a marketing stunt, investor bait, or “MrBeast‑style” spectacle rather than serious research.
- Others defend it as useful exploratory R&D to reveal AI failure modes before such systems are deployed less responsibly.
- Underlying tension: is this helping society prepare for an inevitable future, or actively constructing a dystopia while claiming reluctance?