You’ve heard the iconic phrase. In the 1983 film “Wargames”, high school student David Lightman (Matthew Broderick) unwittingly hacks into a military supercomputer while searching for a new video game to play. He thinks he’s playing a game called “Global Thermonuclear War” - - the computer takes the simulated threat for real. What follows is gripping drama, as Lightman and girlfriend (Ally Sheedy) fight to stop the onset of WW3.
IF YOU ARE GOING TO PLAY A GAME, KNOW THE GAME YOU ARE PLAYING AND KNOW WHOM YOU ARE PLAYING
On Saturday, February 22nd, President Trump praised Elon's efforts and then prodded,
“Elon is doing a great job, but I would like to see him get more aggressive. Remember, we have a country to save, but ultimately, to make greater than ever before.” - Pres. Trump, on Truth social
Musk relied shortly thereafter on X, saying "Will do, Mr. President!" And then, this:
Musk's statement was followed up on Saturday afternoon, with an email from the Office of Personnel Management with the subject line, “What did you do last week?”
The email body stated, “Please reply to this email with approx. 5 bullets of what you accomplished this week and cc your manager.” The email also told employees to respond by midnight on Monday and not to include classified information.
The email has caused quite a stir the past four days. There have been a wide range of responses, including those arguing that it is dumb, that Musk lacks authority, that it breaks protocol, and on and on. Some senior officials instructed their staff not to respond or to wait for further guidance, Kash Patel at F.B.I., for instance.
There is a lot of scuttlebutt around the move, but I'd like to delve a bit deeper, to see if we can learn from it.
In a previous post, "Sticks? All out. Carrots? Running low. Getting people to move.", I shared Professor Clayton Christensen's admonition to my HBS section: Rather than saying that a decision maker is "stupid", can you figure out why and how a decent person might reason themselves into that decision?
Why would Musk and crew make this move? What is the game?
MUSK KNOWS DOGE HAS TO PRUNE, AND PRUNE QUICKLY
Musk and DOGE have been tasked with improving the efficiency of the Federal government, and the first task at hand is to prune, to cut waste. His task is enormous, and time is running out.
“The Federal Register lists 438 Federal government agencies and sub-agencies. There are about 24 million government employees out of about 170 million workers in the US, or about one in seven workers. … Government spending is at World War II levels, as a share of the US economy. … Resulting in an astronomically-large, public debt. … The Penn Wharton Budget Model (PWBM) believes the US has until about 2040 to correct the issue before catastrophic default …”
- Semper Doctrina, "Peter, Parkinson, Praying For a Benevolent Dictator"
ELON DID A SUCCESSFUL “REJUVENTION PRUNE” AT TWITTER
“… Elon Musk fired 80 percent of the Twitter staff when he bought the firm in 2022. Two years later, it's now called "X" and has some added product offerings like an AI, "grok". Today, the firm operates with less than 40 percent of the staffing level of Twitter. It begs the question, what did all of those other people do, exactly?”
- Semper Doctrina, "DEEPLY SEEKING DISRUPTION. The Tech Empire's New Clothes"
[in a dire situation] … "rejuvenation pruning"… involves removing all the branches, all the way down to a few inches from the plant’s root ball. This is done when regular pruning can no longer solve the problem. It is grave, and the plant often doesn’t survive.
- Semper Doctrina, "HEALTHY PRUNES. Just a trim or to the root ball?"
Musk was successful in his rejuvenation prune, wildly so. He followed that up with the stellar success of his GrokAI, leapfrogging his competition at hyperspeed. Musk comes to the DOGE challenge with a wealth of experience of moving boldly AND quickly at SpaceX, Tesla, xAI, Neuralink, Boring, and more.
DOGE MUST PRUNE MORE SURGICALLY WITH FEDERAL GOVERNMENT
To prune surgically, the gardener must closely examine the whole tree, make the easy cuts, and then choose wisely when making the more difficult ones.
“Let’s start with the easiest cuts to make: The "Three Ds," dead, diseased, or damaged. …
The next three pruning commands are more difficult, mostly because they remove perfectly-healthy, strong branches - - branches whose very strength and vitality impedes the growth YOU want. Branches that cross, are acute, or have clusters must also be pruned. … “
- Semper Doctrina, "HEALTHY PRUNES. Just a trim or to the root ball?"
MUSK’S MOVES ARE DESIGNED TO QUICKLY MAKE THE EASY CUTS
It is important to remember that this week’s email comes on the heels of the Fork in the Road program email on January 28, 2025 presenting a deferred resignation offer.
More than 75,000 federal employees accepted that offer and the eight months of severance that came with it. Another 220,000 federal workers were in the category “probational”, those with less than one year on the job, and are the focus of agency personnel cuts. These two groups total nearly 300,000 federal workers, or about 10 percent of the three million overall count.
Thus, ten percent of the federal staff respresented very obvious, easy prunes, easy cuts.
You know what else is an easy cut? Dead people! Or, non-existent people.
While we don’t know yet what has been caught in this “net”, what if it is material?
MUSK’S RECENT EMAIL REQUEST, “WHAT DID YOU DO LAST WEEK?” COLLECTS INFORMATION TO INFORM MORE DIFFICULT CUTS
I wrote a post shortly after the election, covering DOGE in some detail. In the post I stated that Trump’s Musk, has something that Kennedy’s McNamara and Reagan’s Grace never had, A MICROPHONE. That is, I said, “They never had a mass-communication media like the one Musk controls.” I argued that Musk and DOGE can both crowdsource solutions and shame the recalcitrant into making changes.
And, indeed they have. If you look at social media commentary, on X and elsewhere, you will see this happening.
It was reported today that one in three federal workers responded to the email request by the Monday deadline. That is, one million out of the estimated three million federal employees.
I actually participated in it myself. It took me about a minute and a half,” - White House Press Secretary Karoline Leavitt
“What he’s doing is saying, ‘Are you actually working?’”, “And then, if you don’t answer, like, you’re sort of semi-fired or you’re fired, because a lot of people aren’t answering because they don’t even exist.” - Pres. Trump, Oval Office meeting with French President Emmanuel Macron
You know, come to think of it, Trump’s Musk, has one more thing that Kennedy’s McNamara and Reagan’s Grace never had, the WORLD’S BEST AI.
THREE MILLION PEOPLE JUST FED A LARGE LANGUAGE MODEL (LLM)
If you are a federal worker who received the email, your choices were pretty simple.
Perform the menial task in a response by the deadline - - > 1,000,000
Protest or object to the task in a response by the deadline - - > ????
Fail to resond in any manner by the deadline - - > ????
We could parse it more: The F.B.I. people and others who followed leadership instructions to withhold a response, are in another bucket, for sure. But, even that bucket is in the LLM.
Large Learning Models, or LLMs, are a type of AI algorithm that is extremely good at understanding human language and text. When used for “sentiment analysis”, they can detect nuances, capture subtle expressions, and understand emotions. You, yourself, have likely interacted with a customer service chatbot LLM.
I ASKED GROK WHAT IT COULD DO IF IT READ THE MILLION EMAILS
SD: “Could you tell me what types of things you could learn if you were to read the 1,000,000 responses from federal workers who submitted answers to the “What did you do last week?” Email from opm.gov?”
Grok: “You’d get a broad snapshot of daily tasks and priorities. … Patterns in workload and productivity would stand out too. … You’d also pick up on the tone and morale. … Bigger-picture trends could surface too—like how much remote versus in-person work … Of course, the data wouldn’t be perfect. With a tight deadline and mixed guidance, not everyone would respond fully or honestly. Some might exaggerate to look good, others might downplay sensitive work.” [image below is full text]
SD: “Tell me about what you would learn from sentiment analysis of the email responses”
Grok: “… fascinating window into how they’re feeling about their jobs, their agencies, and maybe even the government as a whole. Sentiment analysis would break down the emotional tone—positive, negative, neutral, or mixed—based on word choice, phrasing, and context, giving you a pulse on morale and engagement without needing to read every reply manually … The analysis wouldn’t just tally happy versus grumpy—it’d reveal intensity. Strong negatives (“this place is a dumpster fire”) versus mild ones (“eh, same old grind”) would flag where discontent’s boiling over. Same with positives—gushing “best week yet!” versus a tepid “it was fine” could separate genuine enthusiasm from going through the motions. And mixed sentiments, like “killed it on the report but hated the process,” would expose nuanced trade-offs in day-to-day work.” [image below is full text]
BUT SEMPER DOCTRINA, WHAT ABOUT THE TWO MILLION ABSENT?
SD: “What might the responses allow you to learn about the non-responsive?”
Grok: “… you’d spot who’s conspicuously absent by agency or role … If sentiment among responders tilts negative—like “this email’s a waste of time” from IRS clerks or “leadership’s clueless” from Education staff—the non-responsive could be the extreme end of that: fed up enough to ghost the request entirely. … the non-responsive might be disconnected, understaffed, or just off the radar. …” [image below is full text]
As a final note to the discussion, I asked Grok how long it would take to read the million emails and do various analyses, like sentiment. Answer: “hours at most for a thorough rundown”.
NEW GAMEPLAYER IN DC IS AI - - NEW GAMEPIECES ARE DATASETS
What we have covered thus far, is child’s play, trivial for an AI like Grok. Just imagine the other data sets it has access to, public and private, for individuals, for groups, and for organizations. Imagine the analyses it can do and do nearly instantaneously, finding patterns, finding linkages, suggesting courses of action.
What would I learn if I asked Grok which positions seem unimportant? Redundant? Easily outsourced or automated?
Would Grok, the AI gardener, be able to spot those branches that are “crossed”, “acute”, or “clustered”? Something tells me it can. Something tells me it will.