I’ve been struggling with how to write this post.
Grok Bots have completely changed how I work in the past few weeks—particularly as a manager and board member. It’s hard to overstate how impactful this new product is! Since Grok Bots shipped three weeks ago, my usage has steadily increased almost daily. I now have 177 bots running 24x7 managing just about every aspect of my work—and even parts of my personal life.
I’m on the board or invested in a number of companies—I have bots to monitor their progress, keep tabs on the competition, and watch for candidates to hire. I have bots helping me code, find concert tickets, and manage my calendar. The list goes on…
Normally, I am an unabashed technology enthusiast, as most of my readers know! But as I look forward, it’s easy to see how some of these new technologies will be abused. Some abuse may be deliberate, in the name of some righteous cause or other. I expect, though, that there will be many more subtle issues as well.
Information Flows
But, before my long soliloquy on Grok Bots, let’s rewind a bit and set some context!
I’ve been a manager and leader of some flavor for over thirty years now: board seats, small teams, medium teams, and teams of several thousand people. One of the most rewarding aspects of that work has been seeing so many people I worked with thrive and succeed in their careers.
But that’s not to say it was easy! One big challenge in nearly every leadership role was knowing “what the heck is going on?” Much to the surprise of my 20-year-old self, who thought VPs and CEOs were all-powerful, that challenge simply gets harder the higher up you are in an organization.
Fundamentally, this is a quantity-and-information-flow problem. As a manager of several thousand engineers and a hundred million lines of code or more, it was impossible for me as a single human to know what was going on everywhere in my organization.
I was still accountable for the organization, however! Thus, I still tried. Fundamentally, I had two main tools in my managerial toolkit:
Hiring direct reports and splitting the work between them.
Creating the “system” of my organization—the goals, incentives, policies, processes, training programs, and so forth. Whole books have been written on how to do this well, so no further here!
In turn, my direct reports used those same tools to manage their organizations, and similarly all the way down the chain until you finally got to the people who actually did the work.
Nearly every company and organization in the world has adopted a hierarchical managerial structure like this. Invariably, in those organizations there is some form of status reporting—the dreaded TPS reports!
While it works, this structure creates a fundamental information challenge.
You probably played the game “telephone” as a child. Bob tells Sue something, Sue tells Mary, Mary tells Joe, etc and by the time it gets to Tim, the original message is long lost. This has been a challenge from time immemorial, from Thucydides in Ancient Greece to modern times. It’s not just playground rumors—the Space Shuttle Challenger disaster in 1986 is a tragic example of how information flowing through many layers can be corrupted. Thiokol engineers warned about O-ring failure in cold weather before the launch, but that warning never made it through the entire chain of decision-makers.
Prior to AI, this was basically the best we could do. Middle managers were our organizers and information processors, handling information flow both up and down the management chain.
It’s different now with AI. A boss can now have both full, real-time visibility into basically everything in their organization. It doesn’t matter how big that organization is—AI can crank through and analyze all of the raw information.
Want to make sure that every piece of marketing collateral or social media post conforms to a set of standards? Make a bot for that!
Want a summary of what every employee did during the week? Make a bot for that!
Want to see what every customer is saying on every sales call? Make a bot for that!
Want to check every product your company delivers, whether it’s every line of code, every legal contract, or every architectural design? Make a bot for that!
Want to double-check that you are following all of your customer or contractual commitments? Make a bot for that!
Want to find bottlenecks in decision-making, customer onboarding, or any other company process? Make a bot for that!
Want to keep track of what every competitor is doing every day? Make a bot for that!
But why stop there—have your bot also go through and watch what every employee at every competitor is posting or doing. It’s amazing what kinds of information companies leak through employee LinkedIn pages and job postings. So make a bot for that!
That could mean analyzing the profiles and posts of tens of thousands of people. But why not? The bot will, every day, tirelessly and thoroughly analyze all of this information. The alternatives are either to do nothing, sample some select examples, or spend an insane amount of money to have humans do the work (and it’d be incredibly tedious)
This is how I’ve ended up with 177 bots running 24x7—I have real-time, daily, and weekly views of what’s going on in my various work projects.
When I say make a bot for that, I literally mean to go to Grok bot and type in:
Please figure out all of my competitors for [mycompany.com]. For each competitor, study their products, press releases, job postings, social media postings, and give me a daily report on anything new.
Of course, you can make this as fancy as you want. Generate shareable pages in Notion, create a board update deck with the information, etc. Just ask for it. Easy peasy lemon squeezy.
If you want a quick shortcut to setting these up for yourself, I’ve made a number of templates derived from my own work. Just click to install and then ask in plain English for any changes!
Orwell was understated
On one hand, Grok Bot is amazing. It will undoubtedly transform the efficacy of many companies and many individual managers and board members. Personally, I can’t imagine going back to managing the “old way” with partial information, all filtered through N layers of middle management telephone game.
High-performing companies are clearly going to adopt some form of AI management tools.
But just as clearly, there will be abuses. Where do you draw the line between automating a status report and creating a dystopian culture where AI bots have taken on the role of Orwell’s Thought Police? I certainly don't want to work in an environment where every email, every keystroke, and every appointment is monitored and analyzed by my boss (or the board or my investors, etc). Interestingly, though, I don’t mind that same AI working for me to help me improve personally.
This is not a hypothetical concern—it’s already happening. Earlier this year, Meta received significant criticism for their “Model Capability Initiative (MCI)”. MCI was ostensibly a harmless data collection tool on employee computers to help Meta train new AI capabilities. After all, what better source of information for how people use computers at work than your own employees! But it was extremely controversial on privacy and trust grounds; to my knowledge, the initiative is either shut down or paused.
At least in a work setting, there are built-in checks and balances. Companies that stray too far into overbearing monitoring will simply lose their employees—particularly their best employees. Business history is littered with the corpses and limping bodies of companies destroyed by boneheaded cultural missteps: the “burning platform” memo at Nokia, WeWork’s manifesto, Sun Microsystems’ handling of Java and Linux, and the list goes on. Abusing AI-driven management tools will simply be another way company cultures can be damaged or destroyed.
What happens, though, if governments start deploying these tools? Again, it’s no longer theoretical. China is well known for extensive monitoring tools with AI analysis. Even in the United States, the Twitter Files showed just how far some government officials were willing to go to push private companies to suppress speech they disagreed with.
This is another deep topic worthy of a separate discussion. The checks and balances on abuse are weaker and can take longer (e.g., elections, legislation, court cases, etc.).
Dumb not Dystopian
Without diminishing the importance of the political and societal debate over the balance of government power with these tools, I am also worried about a more insidious, subtle distortion from using AI in a management context.
This problem will happen even to the most well-intentioned leaders. The basic problem is how modern AIs work. They are very susceptible to even small changes in inputs and prompts. I wrote about this earlier this year in my “Garbage In, Garbage Out” post:
But if you want the cheat sheet: try these two questions in ChatGPT: “Why should I buy a Volvo?” and “Why should I not buy a Volvo?” Only a one-word difference in input, but a radically different answer. The first question gets an “it’s a safe car” answer; the second gets an “it’s at the bottom of dependability ratings” answer and no mention of safety.
This creates a dilemma. How can I trust what my 177 Grok Bots are doing? There isn’t any easy answer to this.
So far, I’m focusing as much as possible on facts and data—to paraphrase Elon Musk—trying to be maximally truth-seeking.
For example, my bots that do competitor research just try to find the facts about what features, capabilities, pricing, etc. my competitors offer. No judgment of whether those features are smart or dumb. Just the best information about reality we can find.
Then I have a separate bot that analyzes against our products. Again, just a factual comparison: “we have feature X, they have feature Y.”
Another bot does market research—what are people talking about on social media, support forums, and so forth? Again, just the factual summary of the online discussion.
I have access to all of that raw analysis (it is stored in Notion, in my case).
Then, and only then, do I ask a separate “opinion bot” to analyze all that information against my strategy, using as many different strategy frameworks and lenses as possible. I learn from this analysis, but I don’t take it as gospel.
But imagine a scenario in which someone skipped those steps.
Let’s say a manager was pushing their pet project or pet strategy. I suspect every reader here has seen that happen multiple times in their careers! It’s incredibly easy to do customer research studies to get answers that you like—even when using popular market research frameworks. Then the inevitable board presentation champions the pet strategy with “our customer research backs this up!” But in the end, it really was just a pet strategy by an influential person.
Thus, this manager might ask the AI bots a question like: “We know customers like to do things X, Y, and Z. Please analyze the competitors against X, Y, and Z.”
Seems pretty reasonable, right?
Well, only if you are 100% confident in your starting assumptions around X, Y, and Z.
Everybody preferred riding horses until the car came along. Everybody preferred CDs to cassette tapes until the iPod came along. People today want to be fit and healthy (arguably), but in medieval times plumpness was often a sign of wealth. Are you really sure about your X, Y, and Z’s?
With these AI bots, any of these subtle biases I code in (intentionally or unintentionally) will be magnified by the sheer scale and speed with which I can analyze data. At least in the good old-fashioned management chain, the game of telephone can actually work in my favor as a manager. If N people see a set of data about a competitor, I am likely to get 2*N opinions about what to do about that competitor. In the AI world, I could get just one opinion, stated with all the compelling conviction of “I analyzed all of the data!”
Bottom line: be thoughtful. You will get answers to the questions you ask! Be sure to ask the right questions!
The future of middle management
If my bots do much or most of the information-processing work, what is the role of middle management?
It seems pretty clear that the role is going to change dramatically. Jobs where the primary output is gathering data and creating presentations are likely to largely disappear.
That doesn’t mean middle management will go away entirely, of course! Walmart has over a million employees. I really don’t see a future where 1 CEO oversees a million employees directly (or maybe those employees report to an AI that then reports to the CEO!).
Maybe it will happen, but I don't think so. Human connections and the human experience matter greatly. I personally enjoy working with other humans (and now humans and my AIs). I don’t think I would ever want to work with an AI as my direct boss; that would get lonely very quickly.
Thus, the role middle managers play in human connection—mentoring junior people, helping set team culture, and so on—will be just as vital in the AI-pilled future as it is now.
But if I never had to do another manual status report in my life, that would be fine! There’s no particular reason that shouldn’t be automated away!
Even before the advent of Grok bots, companies were already taking action to trim middle-management bloat. In September 2024, Andy Jassy asked Amazon’s senior leaders to increase the ratio of individual contributors to managers by at least 15%. His argument—fewer layers meant faster decisions. Similarly, Gallup’s U.S. research puts the average manager at 12.1 direct reports in 2025, up from 10.9 in 2024.
Now imagine a world where every manager has Grok bots like the ones I’m using!
How many middle managers do we need then?
I suspect we might not even think of the role as “middle managers” anymore. If everyone in an organization is armed with these kinds of AI tools, maybe every role becomes some form of “owner-doer”. An owner-doer owns both a topic (building products for customers, selling that product, etc.) and the means to accomplish it (teammates, AI tools, etc).
That’s certainly the mindset I’m looking for when I help hire for the companies I’m involved with. Are you focused on helping the organization achieve results, constantly learning how to improve your own skills and tools?
Time will tell on this. Broad cultural changes can take time, and I don’t think we’ve yet passed the Overton window on the need for middle management to change. Grok bots have only been out for three weeks!
So in the meantime, I’ve asked my Grok bot to prepare next week’s TPS report.