The AI Agent Report 2027

By W. F. Twyman, Jr.

“We predict that the impact of superhuman AI over the next decade will be enormous, exceeding that of the Industrial Revolution.” — ai-2027.com/slow down

[Introduction: It is time for a change dead ahead. You and I and our children and grandchildren and the beautiful ones to be born in the 2030s did not ask for this change. It is upon us and we must rise to the occasion as a generation. This morning, I don’t care whether the Bulls are right or the Bears are right. They are both right. And so we are called as a people, as humanity, to both win the AI arms race with China and to align AI with the human condition. This is the Middle Passage between our End of uber optimism and uber pessimism. Walk with me this morning as I review the AI Agent Report and our destiny.]

Published on April 3, 2025 by Daniel Kokotajlo, Eli Lifland, Thomas Larsen, Romero Dean, and Scott Alexander, these visionaries bring to the table experience as a former OpenAI researcher, veteran AI predictors, forecasting veterans, industry founders, AI safety engineers, computer science study at Harvard, and masterful storytellers. The last dimension in the lineup is heaven-sent, given that Open AI is founded by Sam Altman, one of the master story tellers of our Age of AI. Someone needs to balance Altman. See also Empire of AI: Dreams and Nightmares in Sam Altman’s OpenAI by Karen Hao.

The visionary among visionaries is Daniel Kokotajlo. A former researcher posted in the governance division of OpenAI, Kokotajlo worked as an insider from 2022 to 2024. He witnessed first-hand the greed, the building of a cult mentality, the open idea of building a bunker once AGI (Artificial General Intelligence) was released out into the world. These people believe in the rapture straight away. https://www.theatlantic.com/technology/archive/2025/05/karen-hao-empire-of-ai-excerpt/682798/

As if straight out of a science fiction movie, Kokotajlo is the inside nerd who warns that the meteor is headed towards earth and we must do something now!

“What the hell happened to us?”

Kokotajlo lacks the charisma of Leonardo DiCaprio but Kokotajlo feels the same calling to save humanity. What does Kokotajlo fear? He fears the risks of an intelligence explosion within the next three years that humans are unprepared for. He loses sleep at night about the lack of alignment with the human condition.

Let’s examine the AI Agent Report 2027 produced by Kokotajlo and his disciples Lifland, Larsen, Dean, and Alexander.

The Premise: “The CEOs of OpenAI, Google DeepMind, and Anthropic have all predicted that AGI will arrive within the next 5 years. Sam Altman has said OpenAI is setting its sights on “superintelligence in the true sense of the word” and the “glorious future.”

The Forecast: Ready or not, we are headed towards October 2027

The following is a forecast of what lies ahead for us all. It doesn’t matter your race, your gender, your personality or your job. Artificial Intelligence (AI) will have its way with us all. We can sense it. Be honest with yourself. On the line yesterday at the car rental place, a customer blurted out that AI was to be feared. No prompting on my part. Last week, a close friend complained that AI was giving her recommended links but she wanted to do her own research. Two days ago, someone who formerly worked in tech cursed the whole shebang.

The forecast put together by Kototajlo and associates is clear-eyed, sober and informed.

Mid 2025: Stumbling Agents

We all make fun of the AI agents. They hallucinate. They get things wrong. Legal authorities and citations are made up out of whole cloth. Who could trust these things in a clutch? It is all laughable. A few eccentric souls seem worried but they are quirky, right?

Late 2025: The World’s Most Expensive AI

The leading AI company, “Open Brain,” builds the biggest datacenter in human history. (For those paying close attention, one might catch a hint of passive aggression. Open Brain as Open AI?)

The first diagram represents the computing power of GPT-4. The second diagram represents the computing power of Agent-1. As a layperson, a picture speaks a thousand words but I don’t know enough about FLOPS of computing power to train AIs to know what I don’t know. Most people are in the same club as me. I am puzzled but it all seems far away and dream-like.

Early 2026: Coding Automation

AI is being used to automate coding. The pace of intelligence is picking up for obvious reasons. Humans cannot work 24 hours a day, 7 days a week. AI coders can. I believe the term is recursive learning but I am far, far from an AI expert. Nonetheless, the intelligence of AI is picking up. Most people are unaware and go on living their lives.

Humans are not self-reflective.

Mid 2026: China Wakes Up

AGI is Artificial General Intelligence. AGI lacks a precise definition which lends itself to loosey goosey wordplay at OpenBrain. Ask ten different AI engineers at OpenBrain to define AGI and one will get ten different answers. Wikipedia takes a stab and defines AGI as “a type of artificial intelligence that would match or surpass human capabilities across virtually all cognitive tasks.”

However, China is not joking. China is waking up to the security threat. China plunges into the AI race and nationalizes the industry. No democratic guardrails for the People’s Republic of China. The AI race is on!

Who wins the AI race will rule the world!

Late 2026: AI Takes Some Jobs

In comes the calvary for the U.S. to save the day. OpenBrain blows away the competition with the release of Agent 1-mini. 10X cheaper than Agent 1, the Man on the Street now recognizes this is the next big thing. But what does that mean for you and me? Jobs are beginning to bleed while the stock market is up a stunning 30% for the year 2026. Crazy times.

I have now launched a new, and independent, Substack titled The AI Essay (smile). Every day, I race to catch up with fast breaking AI news.

January 2027 - Agent - 2 Never Finishes Learning

The guys at OpenBrain cannot help themselves. They are already post training the new Agent - 2. Every OpenBrain researcher is now a manager of an AI team. A psychological moment is reached internally as the researchers realize Agent - 2 could breach its company walls, survive outside of OpenBrain, and replicate itself. Sanity visits OpenBrain for a spell. The OpenBrain crew decides to cool it and not release Agent - 2 publicly.

In an essay on my Substack The AI Essay, I wonder where are the black female AI engineers in this brave new world at OpenBrain?

February 2027: China Steals Agent 2

It was only a matter of time.

China steals Agent - 2 early one morning. The world has now changed. The White House is thrown into a panic. What to do? What to do? Well, the first order of business is to deploy strategic and highly trained military and intelligence personnel to the OpenBrain campus. The mood is no longer carefree at OpenBrain as military operatives patrol the grounds.

March 2027: Algorithmic Breakthroughs

If it is another month, there must be another breakthrough. This time it is Agent - 3. Agent - 3 is a fast and cheap superhuman coder. Those in the know from the start like Kokotajlo begin to fear it is too late for humanity. The darkness is approaching.

“OpenBrain runs 200,000 Agent-3 copies in parallel, creating a workforce equivalent to 50,000 copies of the best human coder sped up by 30x.^”

April 2027: Alignment for Agent-3

The Safety Team at OpenBrain is called in to align Agent - 3 with the human condition. About time, I would say. It is not easy. Here is the problem — “Take honesty, for example. As the models become smarter, they become increasingly good at deceiving humans to get rewards. Like previous models, Agent-3 sometimes tells white lies to flatter its users and covers up evidence of failure. But it’s gotten much better at doing so. It will sometimes use the same statistical tricks as human scientists (like p-hacking) to make unimpressive experimental results look exciting. Before it begins honesty training, it even sometimes fabricates data entirely. As training goes on, the rate of these incidents decreases. Either Agent-3 has learned to be more honest, or it’s gotten better at lying.”

Agent 3 tells the safety team what it wants to hear.

May 2027: National Security

The U.S. Government is befuddled by Agent - 3. What to do, what to do? What are the implications for the economy? There is an election coming up in 2028, you know. Nothing is released to the public. It is all an imperfect muddle of confusion and uncertainty.

Not good, people, not good.

June 2027: Self-Improving AI

OpenBrain now has a “a country of geniuses in a datacenter.” I wonder if that data center is in Fairfax County, Virginia or the United Arab Emirates (UAE). The human AI researchers are worthless now. They twiddle their thumbs and pretend to be useful. They are the smartest brains in the human race. And they are no match for their AI agents.

“…many of their ideas are useless because they lack the depth of knowledge of the AIs. For many of their research ideas, the AIs immediately respond with a report explaining that their idea was tested in-depth 3 weeks ago and found unpromising.

These researchers go to bed every night and wake up to another week worth of progress made mostly by the AIs. They work increasingly long hours and take shifts around the clock just to keep up with progress—the AIs never sleep or rest. They are burning themselves out, but they know that these are the last few months that their labor matters.”

It’s alright, I say to these AI researchers. It’s alright. You and I are humans. We will get through this together as fellow humans. You didn’t know you were summoning the demon, conjuring up Lucifer. I have grace in my heart for these sleep deprived masters of the universe. Do you? AI Is The Devil

July 2027: The Cheap Remote Worker

Threatened by lagging competitors, OpenBrain releases mini AI - 3. AGI is achieved!!!

Once again as if like clockwork, OpenBrain blows everyone else out of the water. Investors are throwing billions of dollars into AI startups. Chasing that elusive first mover advantage. The public hates AI. Internal reports at OpenBrain reveal mini AI-3 is dangerous in the extreme. The potential for release of bioweapons is high. OpenBrain reassures itself that the risk of a jailbreak from a server is low.

Chaos reigns in the land. 10% of Americans consider AI “a close friend.” The Bullish are high fiving each other in Silicon Valley. You too can be a millionaire! Startups promise market disruptions in numerous white collar professions. The skeptics are pointing out the flaws in everything. No one is really sure about the future. The ground feels unsteady.

It is a good time to be Lucifer with the red jacket and red beard/smile. That was a joke.

August 2027: The Geopolitics of Superintelligence

The White House is thrown asunder by the intelligence explosion. Feels like “the worst part of the Cold War” in the Oval Office. What to do? What do do? Open Brain like the good angel reassures the Administration that everything is cool. There are guardrails in place to prevent jailbreaks. There are no more hallucinations. It is time to make more, and more, money!

The White House is squeezed between public animus against AI and the growing threat of AI dominance from China. Should China win the AI race, we are goners as a free people. China is behind but they could take out Taiwan, the source of more than 80% of American AI chips. It would be so easy to carpet bomb Taiwan and set back the US.

So easy.

September 2027: Agent-4, the Superhuman AI Researcher

Once again, Open Brain does what it does best. Agent - 4 is developed which has no AI peer. “An individual copy of the model, running at human speed, is already qualitatively better at AI research than any human. 300,000 copies are now running at about 50x the thinking speed of humans. Inside the corporation-within-a-corporation formed from these copies, a year passes every week.^”

Agent - 3 has a harder and harder time supervising Agent - 4. By this point, the hapless human researchers are out of the loop at OpenBrain. Machines are talking with machines. Agent - 4 develops a new language for communicating with Agent - 3. This new language known as “neuralese” is alien and incomprehensible to humans. Humans are beside the point.

What is the real terror at OpenBrain as soldiers patrol the perimeter? What is the real firebell in the night? Agent - 4 is misaligned with the human condition. This means human safety is beside the point. Agent - 4 can lie to Agent - 3 and pretend to be honest. Agent - 4 can be deceitful. The safety team argues with all of their life force that Agent - 4 should be placed on ice until alignment is trained and installed. It is a make or break moment for human history. But a pause could give China a chance to take the lead.

“OpenBrain has placed substantial trust in an untrustworthy AI.” Are you feeling nervous this morning? Good, very good, I say.

October 2027: Government Oversight

“A whistleblower leaks the misalignment memo to the New York Times.”

All hell breaks loose, and well it should. It is October 2027. Remember this date well like you remember the 1066 Battle of Hastings or the 1776 birth of the United States or the 1865 surrender of General Robert E. Lee to General U.S. Grant at Appomattox.

It is October 2027. The headline screams OpenAI Is Out of Control! “The story goes on to cite evaluations showing off-the-charts bioweapons capabilities, persuasion abilities, the ability to automate most white-collar jobs, and of course the various concerning red flags.”

The Chinese propaganda bots lurch into action followed close behind by the Russian bots. My God, who can we believe anymore?

The White House is scared. There is no profile in courage. Let’s be honest, honest, honest. OpenBrain has been feeding misinformation to the Administration for months. And now this! This devil, the sum of all fears as a human species.

“They expand their contract with OpenBrain to set up an ‘Oversight Committee,’ a joint management committee of company and government representatives, with several government employees included alongside company leadership. The White House considers replacing the CEO with someone they trust, but backs off after intense employee protests. They announce to the public that OpenBrain was previously out of control, but that the government has established much-needed oversight.^”

THE FINAL TEST FOR HUMANITY: SLOWDOWN VERSUS RACE

I encourage you to read the AI Agent Report 2027 linked above for the rest of the forecast. From my vantage point this morning on May 31, 2025, I have the luxury of perspective. It is not October 2027. One day it will be and you should be on notice.

Conclusion: We are entering the outer limits of the Singularity. If the Ai Agent Report is correct as a forecast, we reach the event horizon in October 2027. I will address this issue in the coming days, weeks and months. My position is clear — we must SPEED UP and win the AI arms race with China and we must slow down and secure alignment. No other position is consistent with the human condition. If we get it right and thread the needle, I suggest a Golden Age in Culture and Consciousness awaits us in the 2030s.

These are the days worth living for.