When You Work with AI, You’re a Middle Manager.
I got my start in hotels & restaurants, worked my way up to being a hotel general manager, before switching into tech. I’ve been a solo contributor, a team member, a manager, and an outside consultant, at companies big & small.
If you’re reading this, you’re probably used to working in tech with your hands.
You’ve designed, built, and fixed stuff yourself.
You’re finding it less fulfilling to work with AI because you’re discovering that you’re often not doing the designing, building, and fixing yourself, but instead, you’re using plain English commands to tell something else to do it. That something else is getting the dopamine rush of seeing the code compile successfully, of solving a mysterious bug, of putting a finished product into end user hands.
And you hate it.
You miss how your job used to be.
You’re a manager now.
Your job is to use plain English to convince workers to get tasks done in a way you find personally acceptable.
You’re frustrated because the output isn’t as good as you’d like, and you can’t figure out how to get what you really want out of them.
You didn’t hire these workers. They were somebody else’s idea, and you inherited them.
You want workers with better skills and a stronger work ethic, but you’re not allowed to hire the ones you really want. Maybe they’re too expensive, more than your company is willing to pay, or your company has just made it a policy to use workers from a certain contractor.
Some people find that they actually enjoy this new job role, but the important thing to understand is that it’s a different job role than individual contributor or team lead. Management is a specialized career that requires different skill sets. If you’re one of the lucky few who enjoys this and succeeds at it, that’s great! I’m happy for you. For the rest of you, keep reading.
Worse, you’re a middle manager.
You don’t get to pick the strategy of what the company is doing overall. A manager higher than you simply gives you a list of tasks your team needs to achieve, and it’s up to you to convince the team in plain English to get the job done.
You don’t have a budget to give your workers a bonus, and they’re not motivated by free pizzas or the foosball table, and you can’t even threaten to fire them. I mean, you can, but they know you can’t really fire them, so that doesn’t motivate them either.
You start to resent the people above you who set the strategy, set the budget, pick your workers, and hand you tasks.
You start to resent the workers below you who get the dopamine rush and don’t obey your orders.
Middle managers are the ones who get laid off.
When a company gets into financial trouble, middle managers are among the first ones to get the axe. It’s easy for the executives to say to the higher-level managers, “We’re just going to assign the workers directly to you, and you tell them what you want to do. The middle managers were just in the way anyway. This will be faster and more agile.”
To some extent, they’re right.
Sure, we middle managers will exclaim things like, “But we’re the ones who know how to motivate our staff! We’re the last defense against bad quality products! We’re the ones who enforce best practices so that all of the teams work in the same way!
Management just doesn’t care about that. To them, middle management layers gradually become a cost and a burden.
I see the demotivation & layoffs all across tech.
It’s developers, database administrators, engineers, you name it. (In particular, this post was inspired by Eugene Meidinger’s post about the (alleged) death of Power BI developers.)
I’m seeing the frustration, the anger, and of course, the middle management (tech) job cutbacks. The job market for middle managers is only great during times of economic expansion and company growth, and let’s be honest, that’s not what we have right now.
I wrote this to tell you that you’re not alone.
Some of your coworkers are struggling with these feelings too, so now’s a great time to check in on them. If you need an icebreaker, you can send them this post and say, “Hey, are you feeling like this too?” (I’m not trying to go viral with this post or anything – I don’t give a damn if it gets hits or not, because it’ll reach my regular readers and that’s my only goal anyway.) Your fellow middle managers are good people to hang out with in person, to share the miseries of managing your terrible workers, the burdens of dealing with the managers above you, and celebrating the human connections you still have.
Free, 3× a week
Get my new posts by email
Three posts a week, plus a Monday roundup of the best database news from around the web.
Brent, you clearly have a lot of knowledge of SQL Server… but as someone with a knack of explaining complex subjects clearly you really have no equal!
Speaking of PowerBI and is it dead: someone built us a PowerBI front end working from a BigQuery database. A need arose to expose it to outside customers so I checked it. It is horrifically slow, ugly and glitchy. It reminded me of my days as a consultant when I would be handed a pile of cr8p like that, be given limited access credentials and a desk next to the toilets and told to make it perform like a rocketship.
No way was I going to let customers see it. So I rewrote the whole thing in C# and Blazor in a day using Kimi. Despite going to the same data source queries run in seconds. It also looks great. And during the migration process Kimi pointed out and fixed various queries in the PowerBI that were incorrect – so it’s more accurate too.
So people can draw their own conclusions!
Thanks for the kind words!
Step back just a little from your story, and it could be interpreted as, “Version 1 built with Product A by Person I sucked, and now we’ve got a Version 2 built with Product B by Person II. It’s faster, and Person II found bugs in Person I’s work.”
That can be true, while at the same time these other things can also be true:
So just be careful when you word the messages of success, that’s all I’m saying. I don’t doubt that your solution is faster/better/cheaper than the original one – but that doesn’t mean Power BI sucks, or that the original implementer sucked, or that the new solution is bug-free, and so forth.
But I absolutely agree with you that we can build stuff today cheaper and faster than we ever could before. It’s amazingly exciting as a technologist.
It wasn’t so much what they built sucked, it is that the platform itself sucks.
It’s the same exact data source (BigQuery).
So why do PowerBI screens take a minute to populate?
Then you notice that the date range has mysteriously reset itself, so you have to change it, and wait again.
The selling point for tools like PowerBI has always been, well, it’s not as good as a custom app, but anybody can modify it
AI has now blown that selling point out of the water hasn’t it
I recall something you said a couple of years ago concerning AI to the effect of “It will just enable you to do more, just like improved tooling always has”
It will also make a lot of the older tooling obsolete!
Rich – I get that you’re determined to bash Power BI here. I’ve heard you, and I’ve gently suggested a few things to think about. You’re responding by bashing it harder.
This blog post isn’t about whether Power BI is better or worse than what you built by hand. Maybe what you built is fantastic and Power BI is terrible. Maybe you’re holding Power BI wrong. I can’t tell from here, and I’m not interested in having that discussion.
If you’d like to talk about what’s in the post, you’re welcome to comment about that here. If you’re looking to discuss Power BI limitations or capabilities, this isn’t the right blog post, and it’s not the right blog altogether. Hope that’s fair.
(I would also add just to be clear: I’m not saying I disagree with you, or that I agree. I just want to keep the comments focused on the issue I’m talking about in the post.)
You are right of course.
Have a good one!
It’s so funny – I just now recorded an Office Hours episode off the back of the cruise ship, en route to Juneau, and I saw your question in there about Power BI! I laughed and discussed it at length there (because that’s definitely the right place to do it.)
I’m glad you posted it there! Thanks for that.
My apologies, for hijacking your thread!
Enjoy your cruise!
Hi Rich!
Author of the referenced blog post and former Power BI MVP here. Brent has clarified that this isn’t really the place to argue the technical merits of Power BI, so I won’t get out into the weeds.
What I will say is that I’ve seen Power BI be hugely impactful for business users, especially in accounting departments. People that just want to get sh*t done and don’t feel supported by IT.
Will Microsoft’s Fabric Apps replace that and enable even more non-devs to get sh*t done? Maybe! Even if it does though, I will be intensely sad if Power BI dies since in my life I’ve seen the tool be transformational for non-coders.
Thanks for the mention, Brent. I like your framing here because it makes everything feel grounded. Ethan Mollick is always talking how management expertise is now a key skill.
https://www.oneusefulthing.org/p/management-as-ai-superpower
I think it’s also a useful reminder to folks about something anyone who contracts or consults should ask: how can I show I’m a profit center, not a cost center? How can I show I’m adding value with my work is intangible.
I think for a lot of us it’s going to feel like a bit of a barroom brawl, sadly.
My pleasure! Absolutely agreed about the profit center vs a cost center. In the age of $50-$200/mo nearly-unlimited AI plans, it’s increasingly hard to say, “For $XX,XXX per month, I provide the service of you telling me what to do in plain English, and I produce code.” We have to find a different way of marketing our value.
Awesome post Brent, thank you. Needed to hear this after a intense week of mindset-changing work via the MSSQL MCP Server. My flabbers were quite ghasted.
At some point I just realized that the deliverables that AI can do are better for the overall goal of the organization’s progress (and will make me more valuable), and feeding my Power BI ego (by being a bottleneck for report development) only hurts me in the end.
The best I can do is continue to be a good steward of the data itself and give the AI tools as much clear direction and context as I possibly can. Then validate and ship. It’s what I did as a manager of data teams, and the employees themselves are now different.
Really appreciate these words!
You bet, sir, glad you enjoyed it!
I hear the argument that Power BI will let non-IT people like accountants create reports quickly and easily. And I think back to “Access will let non-DBA people create databases quickly and easily” and “Frontpage will let non-developers create web sites quickly and easily.”
And then came loads of Access databases and Frontpage web sites that broke every rule in the IT book. And performed terribly. And IT had to support, because non-IT people think IT has a magic wand to fix stuff.
In truth, Access and Frontpage weren’t actually terrible* if you understood and paid attention to their technical limitations. But those limitations were often documented in IT lingo that the non-IT people could not understand, even if they bothered to read them.
Perhaps AI’s biggest benefit is that it actually has read the documentation.
*I’ll admit “not actually terrible” is a pretty low quality bar.
To me, working with AI is more like Project Management as opposed to middle management. With AI you need to give it a specification and then take responsible for testing. Then provide feedback and further testing.
No upper managment can/want to do that.
I hear you, but … there are a lot of AI users who are not doing what you describe (giving it a specification and taking responsibility for testing.)
Oh Brent, I’ve worked in the tourism sector too before moving to tech as well (tourist guide, airport steward and hotel receptionist).
Nice article as always, I’m going to share it with my Middle Managers 😉
Thanks sir! I started with hotel front desk too, and worked my way up from there.
Hi Brent. As usual, I toughly enjoyed how you present your thoughts. Thank you!
I’ve been in tech for a long time and have had a very successful and blessed career. Companies are profit driven and AI’s promise of reducing cost speeds adoption. Techs need to work so they are incentivized to adopt AI as fast as possible. What are the big picture risks of this rapid adoption? I see these interconnecting risks:
Increased dependency on a relatively small pool of big tech providers.
Reduction in invaluable human capacity (head count as well as intellectual and creative capacity).
From a more philosophical perspective, everyone should have the right to work, express creativity and challenge themselves. None of these rights are part of the corporate equation. It’s easy to say they are, but profit is what drives corporate decision making. Where does this end up 5 to 10 years from now? The train has left the station and can’t be stopped, but since you started this conversation, I figured it was a good time to share my unfiltered thoughts ?
Thanks, glad you enjoyed it.
As far as increased dependency on a small pool of tech providers, I think that’s been the case throughout the history of computing, going back to the days of IBM and Sun. The providers in the pool change, but it’s always been a small pool.
In terms of a reduction in “invaluable” human capacity, I’mma be honest: there are a number of people in tech who are only here to cash checks, and you wouldn’t describe as intellectual or creative capacity. We can debate about what that number is, but if AI can replace grunt work that we would normally have to hire the lowest common denominator to accomplish, and have it done better/faster/cheaper, I’m not sure there’s a downside there. It’s akin to getting rid of punch cards – the people who were punching the cards weren’t necessarily our best and brightest.
About the philosophical perspective about people having the right to work – similar discussions have happened over the last 40 years in the manufacturing industry, which is down roughly 1/3 from its peak: https://www.bls.gov/opub/btn/volume-9/forty-years-of-falling-manufacturing-employment.htm I don’t have any answers on that one only because I don’t spend any time thinking about it. I’m a big believer in stoicism, and part of that involves acknowledging but ignoring things that I can’t change. Am I sad that millions of people have lost their manufacturing jobs, and that entire cities have been decimated by factories moving out? Absolutely. Do I spend time trying to come up with answers for that? No, but I hope folks who actually have the power to change that are spending time thinking about it.
I totally understand that not everybody’s into stoicism – I just wanted to explain why I don’t really engage in that kind of discussion. I’m totally fine with it happening here on the comments, with other people discussing their thoughts, I just wouldn’t chime in since I have no educated opinion on that matter.
Thanks Brent. I appreciate you!
Dependency on a small number of vendors as a problem has always been a myth. If something no longer meets requirements, then you move. The same thing happens with your car and other durable consumer goods. Organizations that purposefully try to fragment their environment just end up working for the environment instead of the environment working for the organization.
Pareto’s law is completely undefeated. The majority portion of most tech organizations are not actually invaluable. Most are resources you just have to have to get the job done, and if you could find more of the top performing 20% you would staff them with those, but you can’t find them. In the restaurant industry we called them “warm bodies;” The lower level staff that maybe you have 4-5 of them combined are as good as your leads, and if you could get more leads you would, but they just aren’t available. It is not a matter of experience or training. Most people just cannot do things that are difficult. They either lack the work ethic, the drive, the innate ability for the domain specific skill set, or (a really big one) don’t have the effort-reward tolerance necessary to get there. A fantastic example of this for me, was during covid. The department I worked in effectively went from 180-200 IT staff to 14. We didn’t have the tools in place to collaborate or work remotely as an organization. We thought we were going to get absolutely killed. Instead, about every 2 months were starting and closing out 1+ year duration projects. We had projects that had been in back log for a decade get implemented and closed. Availability and performance went up. Helpdesk tickets were closed faster than they ever had been before with the added challenge of doing everything 100% remote. All it took was getting 95% of the staff to stay on the clock and off their phones, email and ticketing systems.
AI will not lie about having done something, won’t call out sick, is always on time, won’t deviate from assigned tasks or priorities and if you compare it to a junior, it has a far higher accuracy rate. There is a shortage of tech skill available in the industry that cannot be closed through training. Most people just cannot do it at a high level. Adding more low level staff who will never advance to that high level only serves to increase costs and alienate the seniors responsible for driving the organization forward. As Brent said, most of the industry is not there to practice their trade, they are there for a paycheck only. Or worse, for their ego.
Whenever there is a disruptive technology, whether it be going from manual farm plows, to horse pulled plows, to motorized farm equipment plows, it has always served to free up labor that can be productive doing something else. Or going from office couriers, to phone switchboard operators to PBXes. Or Secretaries to Microsoft Outlook. Or data entry clerks to mini computers. Usually at a higher pay standard that before.
Keith, there are lines in here that are simply not true:
This happens all the time. It’s a running meme amongst those of us who use AI, because it’ll constantly backtrack and apologize, telling us humans that we’re right, that it made a mistake, that it shouldn’t have said something was done when it wasn’t, etc.
AI providers can, and do, have outages.
I literally just got done asking ChatGPT Codex why it stopped working on a particular task when I’d told it to loop until it was finished. I’d left my laptop alone, told it that it needed to be done when I got back. When I returned, it complained that I hadn’t been there to authorize it to use a text editor. I asked if it could edit text files itself, without using a text editor app, and it replied yes, you’re right, I shouldn’t have asked to use an external tool when I’m perfectly capable of combining lines together, sorry about that, just let me go a while longer and I’ll get it done.
I’m starting to think you just haven’t used AI for any meaningful project. LLMs absolutely do go off on tangents. I vividly remember a recent weekend where I returned to the computer to find several dozen pull requests that it had opened for unrelated issues that it thought were important and thought I’d want to have, and still hadn’t finished the original task I assigned it.
Ah yes, you just magically move. How hard can it be to replace things?
Didn’t you read your own advice, Keith? If something doesn’t meet your requirements, you just MOVE, Keith. You said it yourself. Just move.
Wink wink. You’ve got a lot of text in here, but it doesn’t hold up against investigation.