The fantasy football world got a jolt last week when longtime Yahoo fantasy analyst Andy Behrens, along with fellow Yahoo fantasy analyst Dalton Del Don, were let go from their jobs.
Pretty clear from Behrens’ post on X that it came as a shock.
And while he got the typical “aww man, that sucks” and “what are they thinking?!?!” types of replies, it wasn’t until earlier this week when the fantasy community sounded the alarm.
What happened was simple enough: Fellow fantasy analyst Paul Charchian of Fantasy Life posted a Yahoo fantasy story about Detroit Lions wide receiver Jameson Williams and how he was poised to be a superstar.
Not a hot take; many analysts have noted this.
But what caught everyone’s eye was the topper to the article, before the actual “analysis.”
It was italicized so no one could miss it: “This article was written with the assistance of Castmagic, an Al tool, and reviewed by our editorial team to ensure accuracy. Please reach out to us if you notice any mistakes.”
That caused a stir. JJ Zachariason of Late Round Fantasy Football retweeted it, Jake Tribbey of Fantasy Points (and a former Fantasy Sports Writers of America writer of the year) did the same, noting in his post that he was “sad,” and the general sense across all of fantasy sports was pretty much summed up in this response:
All told, the fantasy industry — kinda, sorta, maybe for the first time — got a whiff of what a lot of other industries are staring down the barrel of: Complete and utter disruption, computers taking over, the end of the world as we know it. And it’s all happening soon.
Fact is, they’re not entirely wrong.
And I’m here to underscore something else: “Soon” is more or less today.
Look ma, I built a sim!
And now, a short list of things I am incapable of doing, presented in no particular order: Writing computer code; using computer code to create simulations of the 2025 NFL season; using those simulations to project player performance; using all of that to draft best ball teams; using all of that to draft the “best” best ball teams by testing out different draft strategies; simulating all of this tens of thousands of times; using all of that to suss out what others might be doing when drafting their best ball teams; and about 120 million other things.
By the way, that “120 million other things” is not some fugazi number I pulled from thin air; that’s how many computations ChatGPT did for me earlier today when I asked it to run a detailed simulation of Underdog’s Best Ball Mania VI (BBMVI).
Some context for that number: According to — ahem — ChatGPT, a single Google search is about 1,000 computations. Put another way: All of Photoshop — as in, making Photoshop — would take about 120 million computations.
At first I wanted it to draft millions of teams with millions of ways to go about it. After 16 hours of running in the background, ChatGPT finally came back and said it couldn’t do it.
“Like trying to launch NASA from a Chromebook,” it told me.
So instead we settled on a 10,000 draft simulation, which wasn’t what I was hoping for, but, well …
Yeah. I — me! — just ran a 10,000 draft Underdog simulation on ChatGPT. Took about 45 minutes. All I did was tell it what I wanted, uploaded some rankings, and let the thing go to town.
And to be clear: In my experience using AI, nothing has taken more than a few seconds. The 45 minutes showed the depth of whatever the heck it was doing.
Magic machine
“Any sufficiently advanced technology is indistinguishable from magic,” the famed science fiction writer Arthur C. Clarke wrote in 1962.
No kidding. As I said a few sentences ago, I have no idea — none — what was going on in the background when I asked ChatGPT to run a simulation for BBMVI.
And you know what? I don’t care.
That may sound intellectually incurious, but really: How is it different from anything else I use on a daily basis? I have no idea how this computer I’m typing on works. I have no idea how the car I drive works. I have no idea how the music I listen to beamed off a satellite works. I have no idea how a satellite works. I have no idea how music is recorded in the first place. I mean, records? How do records even work?
We are in serious magic mode when it comes to AI.
Oh, by the way, after I finished up the BBM sim? I fed ChatGPT tonight’s MLB slate on DraftKings, uploaded some projections, told it to scour the web for stats, weather, stadium effects, etc., and spit me out a few GPP lineups. We shall see.
Cause, effect
Now, to be fair, a Yahoo rep told Barrett Sports Media the firings and the use of AI are independent of each other. Here’s the quote:
“Our editorial team has been using an AI tool since the end of February to deliver written summaries of our podcasts. In each instance, these summaries are edited and published by a member of our editorial team and our use of the AI tool is disclosed at the beginning of the article. We will continue to invest in our content team and have exciting plans in place for the upcoming fantasy football season. We look forward to sharing more soon.”
Whether we decide to believe the above or not doesn’t really matter; AI is definitely coming for the jobs of touts, analysts, what have you.
Heck, it’s coming for me: A lot of the writing I do can be done by AI. Rewriting a press release is monkey work for a journalist, but it takes time. AI can knock it out in 10 seconds.
Even deeper dives can be accomplished by AI.
Journalists don’t like talking about it, but even still: According to a Trint study, nearly three in 10 journalists think at least some of their written work will be done by AI in the next year, and only 7% of journos say their organization isn’t using it.
As it pertains to fantasy? Well, plenty of people earn their living poring over numbers, trends, weather, you name it, every day in an effort to give their readers/listeners actionable content.
But what’s going to happen when AI can handle all of that, and more? I mean, it can already, but there still needs to be someone to feed it the right stuff.
I’m sure it won’t need someone to feed it soon enough. Much like it can spit out a high school paper on “The Great Gatsby,” I’m near-100% certain it will be able to spit out optimal lineups for any DFS slate. And offer up +EV bets. And do just about anything else you can imagine. For free. (Or in my case, the $20 a month I signed up for to use this ChatGPT model.)
All of this is uncomfortable, even for someone like me, who’s kinda-sorta waiting for our robot overlords to take over.
Hardcore analysts won’t be needed, or at least certainly not needed in such large numbers; someone engaging to present the information might be all that’s left.
Uncomfortable.
By the way, you know what else is uncomfortable?
What ChatGPT recommends as it pertains to this year’s Best Ball Mania contest.
You want me to do what?!
After looking at projections, after looking at past performance, after looking at different builds, after running a 10,000 draft simulation, after looking at current ADP, after creating an exposure optimization engine (whatever the hell that is), it is telling me that the best bang for the buck, the best chance to win the tournament this year is to employ a … double elite tight end drafting strategy.
A double elite tight end drafting strategy. It was one of the most volatile strategies, but came with the highest ROI.
Now, granted, there are limitations to what it did for me. Short of “building a data center,” as it told me, it wasn’t able to get real-time info on all the players. It didn’t completely take into account past performance, injuries, and the like. It did model most of this, but again: Data limitations.
But also know this: What I did today was impossible to do on ChatGPT (at least the public version at $20 a month) as recently as last year, when I tried.
So who knows what this will look like next year. Or next month. Or tomorrow.
In the meantime, I’ll be the guy drafting Brock Bowers out of the 10 spot … and Trey McBride at 15. Godspeed to me, I guess. Godspeed to everyone who’s job might be affected by this. Godspeed to us all.
