FORMA's
Five Things.

Live And In Person

AI Is Moving At Lightning Speed.
Keeping Up Is Our Job.

One day at the AC Hotel South Bay in El Segundo, built to hand you the latest tools and AI skills you can deploy in your everyday work the moment you walk out. Catered lunch, networking, free premium AI tools, and third party accreditation issued on completion. You leave with ready-built AI workflows, not notes.

Science Knew What Would Work Years Ago.
AI Is What Made It Feasible.

Merck and Moderna announced Wednesday that a personalized cancer treatment worked in a late stage trial of 1,137 melanoma patients. It is the first time a therapy built individually for each person has cleared a trial this size. The idea behind it was never the mystery.

Picture the room. Researchers have long understood that cancer is not one disease, it is your disease. Your tumor carries a set of mutations nobody else on earth has, which means the ideal treatment is not a drug that works for most people. It is a drug built off your own tumor and nobody else's. Somebody says that out loud in a meeting, "what if we could do this for everyone, not just one person," and everyone nods, because it is obviously right and obviously impossible.

That question is the one AI answered. Not how to beat cancer. Doctors and researchers had spent decades on that. The blocker was arithmetic. A single tumor throws off thousands of candidate mutations, only a handful register with your immune system, and each patient gets one shot with room for about 34 targets. Somebody has to score every candidate and call the 34, for every patient, forever. No lab does that by hand. A model does it in an afternoon.

So the answer came back yes. Then AI sat down and waited with everyone else. Merck and Moderna partnered in 2016 and dosed the first patients in 2019. What happened between then and Wednesday was people getting a shot, going home, and living long enough to find out whether their cancer came back. Nothing compresses that.

Worth keeping honest. This is one cancer, melanoma, not all of them. It works alongside an existing drug rather than replacing it. And these are early results from a study still running.

Which is what we would tell anyone asking what AI is good for. It did not find the cure. It answered a feasibility question humans were stuck on, and the trial ran at human speed anyway. Same for what is left. A biopsy, a sequencing run, a batch built for one person, a shipping label. Weeks per patient, every time. But this is good news. Very good news.

Read more →

Who Is The Better Driver, You Or The AI?
The Data Says It's AI, By A Wide Margin.

There is a rumor floating around that when a human driver and a Waymo get into an accident, 99% of the time it is the human driver's fault. We are here to tell you that is bogus. Nobody has published that number. But the real story is close enough that you can see how it got started.

Start with how often these cars crash at all. The Insurance Institute for Highway Safety, which does not work for any of these companies, found driverless vehicles were involved in 68% fewer crashes than human drivers across Los Angeles, Phoenix, San Francisco, and Austin.

Then look at who gets blamed, which is a different question. Claims adjusters decide fault. An objective pair of eyes. Over 3.8 million driverless miles, the cars drew about 3 property damage claims. Human drivers over the same distance would be expected to draw about 12.

Then there is the wait-what stat. Across those same 3.8 million miles, human drivers would be expected to file about 4 claims for injuring another person. The cars filed 0.

Worth saying out loud: these cars drive in mapped, sunny, well lit cities. Nobody has asked one to handle Chicago in February, salt and slush burying every lane line on the road.

One thing to be clear about, because the numbers get mixed up constantly. This is about robotaxis, the ones with nobody in the front seat. It is not about Autopilot or Full Self Driving, where a person is still legally the driver. Those systems account for 56 deaths in the federal database. Different car, different math, different conversation entirely.

So here is the part that does not fit neatly on a chart. These cars have now been involved in three deaths, and been found at fault in none of them.

If you have seen one being tested in a city near you, it is coming. But as impressive as those safety numbers are, the one number they are terrified of is 1. Sitting in the back seat of a driverless car is a real leap of faith, and if you have not done it yet, it is a trip. No pun intended. It has to be right 100% of the time.

Read more →

Three In Four Flyers Are Scared.
They Fly Anyway. AI Is No Different.

A survey this March found that 74% of American air travelers admit to still having anxiety when it comes to air travel. Fear of flying has held steady for decades. But anxiety does not stop them from flying in a tin can at 500 mph, 30,000 feet above Earth.

Hold onto that, because Pew just released five years of data on how Americans feel about AI, and the headline number is 52% more concerned than excited, up from 37% in 2021.

Here is what makes it worth your attention. That number has barely moved in four years. 52, then 51, then 50, then 52 again. Something that stops moving has stopped being a reaction and has become a societal position, like air travel.

So read it like the flying number. It does not predict usage. People who are uneasy about a technology use it constantly, and comfort and adoption are two separate measurements.

But taking a closer look at the numbers under the hood should scare the beejesus out of OpenAI, Anthropic, and everyone in the AI world.

Adults under 30 were the least concerned group in 2021 at 31%. They are now at 55%, the first majority Pew has ever recorded for them, and their excitement cratered from 25% to 11%. Look at social media; under-30s brought this to our world at scale, told us how great it was and forced everyone to join it eventually. With AI, the young generation is saying, "Hold up, something ain't right here." After three years, the under-30 crowd is not too thrilled about where AI is taking us, and that should raise red flags for everyone.

FORMA's Take

Proximity is not endorsement. The people closest to a technology are not automatically embracing and loving it like they did with social media. The group about to spend a forty year career alongside AI is the least excited about it in the country.

If you are rolling this out to your team, the flying number indicates they will use it. But buy-in and usage is wildly different than using it with confidence. They are still going to board the airplane. They are just not thrilled about it. The backlash is getting louder and stronger, and it's coming from the generation AI companies need to address.

Read the research →

Singapore Is Running AI On Human Brain Cells.
Feed It Or It Dies.

On August 17, a data center in Singapore switched on a rack of computers that run on living human brain cells. Not chips designed to imitate neurons. Actual neurons, grown from human stem cells, sitting on a chip and doing the thinking. If this sounds eerily familiar to you then yes, we just described the plot of The Matrix. Ummm, yeah essentially.

Twenty units sit in the rack. Each one holds at least 200,000 living neurons spread across a bed of tiny electrodes. Those cells started out as human blood. Scientists wound the blood cells backward into stem cells, then grew them forward into brain tissue. The electrodes tap the neurons with little jolts of electricity and read the signals that come back. And because these are live cells, they need a life support system pumping in sugar, nutrients, and the right mix of gases. Somebody has to feed this data center. Can everyone now envision when Keanu wakes up and unplugs from the back of his neck? Yeah, that.

Silicon is and will be the only standard for any real foreseeable realistic future. It is faster, more precise, and gives you the same answer twice, and the people who built the brain rack say so out loud. What the neurons have is the electric bill, and we all know what AI is doing to electric bills right now. One unit runs on about 25 watts, a dim light bulb. All twenty together run on less than a thousand, which is a hair dryer. A single Nvidia chip needs around 700 watts by itself. A rack of them can pull more than 100,000, which is about 80 houses.

So this is not a faster computer. It is a wildly cheaper one that does much less, and the bet is that a few narrow problems are worth solving that way.

So two more things to sit with, which will scare the F out of you. The neurons die. Six months, give or take, and you swap in a fresh batch. So how do we figure out ways to keep them alive? And an early backer of Cortical Labs, the company growing them, is the venture arm of the CIA. So what can go wrong?

Well, just rewatch The Matrix. Gulp.

Read more →

The Easiest AI Tool To Adopt
Is The One That Asks Nothing Of You.

Every AI tool you have tried to adopt asked you to change something. Learn how to prompt it. Open a new tab. Remember it exists at the moment you need it. Most of them lost, because a tool that requires a new habit is competing against every habit you already have. That is why these AI companies made their user interface look like a Google search. It is embedded in how we work.

Meeting notes are the exception, and they are the exception that matters most. AI meeting notetakers are the lowest threshold for anyone to deploy and the highest yield for usable output. You turn it on once and then you do nothing differently. You show up to the same call you were going to attend anyway, you talk the way you were going to talk, and the notes write themselves. There is no learning curve because there is nothing to learn.

Adoption surveys in this category are mostly run by the companies selling the tools, so treat any specific number carefully. But the direction is not in dispute. Sit down in a meeting in 2026 and it is highly likely somebody in the room or video call is recording it with an AI meeting notetaker.

Here is the part worth your attention. Almost everyone stops there. The notes get generated, they get skimmed once, and they go sit in a folder with the other forty meetings nobody has opened since. You automated the writing. You did not automate anything that happens after.

The yield is one step past the note. We run our AI meeting notetaker, a combination of Granola, Wispr Flow, and Read AI, for capture. Then we connect our model to it through the connector feature so it can read the transcript directly. Then we add a SKILL markdown file we built that turns that meeting into actionable next steps and clear responsibilities, all through agentic AI triggers.

The flow is short. Add a meeting AI notetaker. Connect it to your LLM. Add a SKILL.md. Then trigger it. One sentence does the whole thing: "run the meeting-os report for the meeting with XX." That is it. Within five minutes you get this.

See the output →

FORMA Tools

Not sure which AI model is actually right for your work?

Six questions. A personalized model recommendation. No benchmark tables, no guesswork. Just a clear answer based on how you actually work.

FORMA Programs

You have learned to talk to AI.
Now learn to build with it.

The next Build With AI session is forming now. Tell us what you want to get your hands on and we will build the session around it.