OpenAI's grand phase-three plan: automated AI researchers by 2028

Back in the 1920s, electricity reached rural America and quietly remade daily life. The lightbulbs were nice, but the real shift came when electric motors, appliances, and radios showed up, eventually unlocking computing and modern medicine.

OpenAI thinks artificial intelligence is right at that inflection point. Co-founders Sam Altman and Jakub Pachocki just laid out their roadmap for Phase 3 of the company, and they put a surprisingly specific date on the wall.

By March 2028, OpenAI wants an automated AI researcher doing a significant fraction of its frontier research side by side with human scientists.

Welcome to Phase 3

OpenAI breaks its history into three distinct chapters.

Phase 1 was pure research toward AGI. Phase 2 started when that work escaped the lab and turned OpenAI into a product powerhouse with ChatGPT.

Phase 3 is all about abundance. The goal now is to take frontier intelligence and make it cheap, safe, and ubiquitous enough to restructure the broader economy.

The plan rests on three pillars. First, build that automated AI researcher within two years. Second, accelerate global economic and scientific productivity. Third, put a personal AGI into the hands of every human on Earth.

OpenAI argues that AI conducting AI research will soon be the single biggest driver of progress. That includes solving alignment itself, where human teams will need AI systems running experiments, catching flaws, and exploring alternatives around the clock.

Drawing the line on human judgment

Despite the aggressive technical goals, the founders took pains to push back against the sci-fi trope of completely automating humanity away.

Altman and Pachocki warned that fully automating human decision-making would be both dangerous and deeply unsatisfying. In their view, AI should help people chase their goals, not take the wheel.

As models grow more capable, they argue the human role actually becomes far more critical. Setting direction, making moral tradeoffs, exercising taste, and deciding what is worth doing in the first place must stay with people.

The memo also warns against power concentrating in a handful of tech giants. OpenAI called for global coordination through an international safety body that could enforce shared standards, and critically, coordinate slowing frontier development down if societal defenses lag behind.

They compared it to the automotive boom: cars only worked for society because we invented seatbelts, driver licenses, crash tests, and road rules.

Why it matters

OpenAI is laying its cards on the table, and the timeline matters more than the philosophy.

For engineers and researchers, the March 2028 target for an automated AI researcher is a warning shot. The bottleneck in AI development will soon shift from human engineering capacity to how well teams steer automated research pipelines. If you build software or models, your value is shifting fast from hands-on execution to taste and high-level judgment.

For the industry, promising a personal AGI to everyone on the planet reveals OpenAI's real endgame. Achieving that scale requires compute and inference costs to drop toward zero, signaling that OpenAI will continue aggressive capital spending on physical infrastructure and energy.

We think OpenAI played this smartly by positioning itself as a defender of human agency and distributed power. They are explicitly addressing public anxiety over tech monopolies before regulators force the issue.

The big catch, of course, is human nature. Proposing an international body that can coordinate hitting the brakes on frontier development sounds great in an essay. But in a brutal global race where commercial and geopolitical stakes are massive, trusting competitors to voluntarily slow down feels like wishful thinking.