OpenAI Exposes 'Polaris' Project — the '2028 Great Unemployment' May Really Be Coming

Polaris: OpenAI's North Star

It has been reported that OpenAI’s chief scientist Jakub Pachocki told MIT Technology Review the company is building a fully autonomous, long-running multi‑agent research system — a project internally dubbed its “North Star” or "Polaris" — with a 2028 horizon. This is not a marketing platitude. Reportedly, the firm has concentrated resources and reorganised teams so this goal overrides other priorities. The timing injects fresh plausibility into a viral “2028 unemployment” article that many dismissed as AI‑generated fearmongering; now the prospect of machines that can run research around the clock no longer sounds purely hypothetical.

Two different bets: OpenAI vs Anthropic

OpenAI has moved by acquisition and internal consolidation — buying developer tool-maker Astral and folding Codex, ChatGPT and browser efforts into a single “super app” — to accelerate a research‑agent future. At the same time, it has been reported that competitors are taking contrasting approaches: Anthropic quietly launched Claude Code Channels to let developers run Claude sessions inside Telegram and Discord, pushing agents into daily engineering workflows today. Former OpenAI researcher Andrej Karpathy reportedly framed this as inevitable: “all LLM frontier labs will do this… it’s the ultimate BOSS fight.” One shop builds autonomous researcher agents; another embeds agents as immediate teammates. Both roads lead to profound productivity shifts.

Safety, economics and geopolitical stakes

Pachocki has reportedly been unusually candid about limitations: OpenAI plans to use other models to monitor agent behaviour, but he acknowledged the current lack of full control — “we don’t understand these models well enough to say the problem is solved.” The economic upside being floated — it has been reported that internal forecasts project roughly $29 billion annual agent revenues by 2029, including high‑price “research agents” — amplifies disruption risks. If machines can run dozens of experiments per night and cost a fraction of a human salary, what happens to the labour market? And what of policy? This race unfolds amid heightened regulatory scrutiny and export controls on advanced chips that will shape who can deploy such systems at scale. Are regulators and societies ready for an era where AI no longer just assists research, but autonomously advances it?