Key Points

• The Player/Coach Mandate: Unlocking high-value AI gains requires combining managerial oversight (setting outcomes and filtering results through judgment) with hands-on individual contributor skill.
• Outcome-Driven Work: Software engineering must shift away from micro-managing task details (which are delegated to AI) to defining and verifying high-level outcomes.
• The Illusion of Competence: AI allows non-experts to “fake it,” but like an untrained spectator stepping onto a professional sports field, their lack of deep domain knowledge will cause them to fail faster and incur major risks (e.g., exposing sensitive data).
• Organizational Friction: Traditional management structures built on micromanagement stifle AI potential. While AI pushers often mistake tool ease-of-use for fast delivery, domain experts often resist AI due to layoff anxiety rather than technical irrelevance.
• The Abstraction Shift: Real adoption isn’t about novelty; it’s about shifting from managing AI as an assistant to letting AI handle the lower-level execution while humans manage abstract systems.

These are my observations building and working with gen-AI for 3 years

I’ve been working on or with “AI” aka ML problems for 6+ years, going ‘all the way back’ to NLP classification and training. Working with gen-AI (which I’ll simply call AI from now on) requires player/coach at every position now. When do you provide managerial context to the AI team and when do you need to step in and roll up your sleeves; you also need exceptional judgment. The models get better but unlocking consistent performance requires a managers skill set along with deep individual contributor knowledge. That’s not to say you can’t be successful without either, however to see the “1000x” improvements some folks claim, you need both skill sets to fully unlock the capabilities. While putting numbers or percentages on AI-ness is non-exact and relatively non-relative, what we’re actually seeing is the inability to accurately express perceived and actual productivity gains, as well as the extended ability aka ‘leverage’ AI can give someone with enough experience in the correct areas. We’re the folks seeing the massive benefit, and living it.

Hopefully AI becomes easier to use, but depending on LLMs to “know” or “judge” has defined limits for things known, aka things that existed during the training. If you’re doing work on the edges of human (or machine) knowledge they can be helpful or a hinderance and player/coach skills are a very hard requirement. If you want to use them to say, create a blog, much like this one 😉, they’ll happily spit out some great code and guide you through the process of hosting.

For software development specifically, there’s an important ways of working shift required. You have to think in outcomes and you have to work in outcomes. Reread that.

Details aka tasks and stories are abstracted to AI. Your goals, well they are co-worked with AI and filtered through your judgment. If you currently struggle with this, it’s the minimum requirement to unlocking “1000x” AI use. If you don’t know what you want and you don’t know how to know that’s been achieved, AI isn’t going to produce ‘the best work.’ It’ll probably give you something ‘almost there’ or fail completly. Even if you have solid outcomes, AI will generally not hit the mark every-time, but it certently get’s you closer to “10x” AI land.

Problem is, most folks are no where near player coach in this space, and it’s easy to think you’re a player when you’ve watched form the sidelines for a long time. That doesn’t mean you can’t learn to play as you go, but to be successful you must know what you don’t know and be able to ask and learn along the way. Your AI coach is going to only get you so far, and from where you started it might feel like you’re now a professional or a super human.

It’d be like walking on to a professional sports team after having no training and wathcing sports for a long time. You probably could learn as you went, but would likely not even be given that opportunity. It would be obvious to everyone around you, that was experieinced, and trained regurally, that you just don’t meet the bar. You might fool some of the other spectators, but it will eventually become obvious as you’ll make more mistakes or just simply not be as good as the others you’re playing with and against. We use to understand meritocracy, but part of the large problem with AI is that’s its very good at fooling folks into thinking how great it is, and how great they are while using it. AI extends “fake it until you make it” however, you become more exposed to things you don’t even know you exist (like putting sensitive data in pubically accessible systems). Fake it until you make it generally fails slowly over time, AI makes it fail much more quickly. Don’t confuse this with learning as you go, or iterative improvement. It’s possible you’re not even a player yet. The good news, it’s never been easier to learn and practice.

For the past several decades, we’ve been treating work increasingly like micromanaging players, we’ve created organizations with extreme managerial oversight to make sure every single task is planned and executed. Coaches have been trained to be player experts and micromanage every move. It’d be like having the coach shoot the basket with you during the game, or holding your hand while you’re at bat. If you need to teach someone, that’s one thing, but a lot of ‘work’ has become folks giving unclear or non-outcome driven demands, and then needing to ‘swoop’ in and fix because others can’t read minds. There is also the perspective of the ‘do it all’ in that folks need to direct and do. That’s not player coach, but that’s what many confuse it to be.

This current way of working does work and it ‘is an approach’ however, those ways of working organizations will never fully unlock a 10th of the potential of what AI can enable. For now, this might be ok, however, someone, or a small group of someones, is going to come with their 10x AI skills or 1000x AI skills and compete very heavily on margin and cost. You still have time to nail outcome driven development and migrate to player coach, but that time window is quickly collapsing.

Some say this will create an AI firing doom loop. Perhaps, however now is the time to work with your humans to build player/coach skills, and change your ways of working. Not everyone wants this. A lot of the AI detractors are highly skilled domain expects, the folks that really know the ins and outs of what they do. We still need these folks, but much like the invention of the type writer, we need to adapt to new technology and get seriously creative how we use it. They tend to frame the criticism of AI through the lens of their experience with their domain, or the apparent use of AI by non-domain experts trying to interface with their domain. The threat of a layoff or firing makes the latter impossible. A lot of the AI pushers, mistake the ease of LLM us with practical building or time to delivery.

While there is an unfathomable amount of hype in the AI space, there’s an increasing divide between people working in the old way, and the people who have adapt to working with AI. The so-called “AI pill “people largely seem to have massive amounts of excitement because they’re doing the thing that was science fiction only a few months ago, but what they’re not expressing or maybe haven’t quite yet internalized is that in order to unlock this incredible capability. I would argue that most of these folks have left ‘working with AI’ and have AI working with them.

You really have to change your approach and adapt to much newer and much more abstract problem handling. That I suppose, is just general advice in 2026. There is no sign of things becoming less abstract. However, if we keep nailing abstractions, many things will become very simple. The ultimate simplicity is dealing with the ultimate complexity.

The beginning of the transformation to things that were considered science fiction a few months ago has already happened, the train is leaving the station, but there’s still time to hop on. We’re not sure where this train is going or how fast it’ll get there, but it is moving and it is speeding up. We’ll need to add rails and build bridges as we go, but eventually we’ll be moving faster then anyone thought possible, and much more so then we are even today.