I think many of us who use AI agents daily -- Claude Code, GitHub Copilot, Codex -- are falling into a trap of thinking "I get it" because we have a positive / impressive experience. I include myself in this.
I'd honestly go so far as to call it cargo cult thinking. This technology is so nascent and changing so fast that not even global experts seem to "get it"; the zeitgeist shifts almost weekly. Trying to keep up.... or worse, projecting mastery to an audience... seems like a fool's game.
What concerns me is the volume of sweeping generalizations I see in conversations and content about how AI works best, or how to approach it (in Power BI, Fabric, generally...). Much of this is well-meant, but it's often very specific to how that person uses AI, agents, their tools... presented as though it's universal. In private conversations, I often get the impression that some people feel they've found some breakthrough in getting agents to work. That's maybe real for them.. but its more a reflection of how potent a coding agent is at enabling someone than it is a transferable (or teachable?) insight. These tools genuinely can help us build impressive, useful things. It doesn't mean we've discovered an eldritch secret and cracked the code.
There's also no widely accepted, objective way to evaluate whether the use of an agent is effective. That's a separate discussion, but it's worth noting, because it means most of these recommendations are based on "vibes", anecdotes, and limited experience... including mine, if I'm frank.
Yes, I'm aware this post is itself a generalization about use of AI and coding agents. I'm not exempt from any of this. Its part of why I've been hesitant to make content about these tools. I regularly catch myself reflecting on a personal experience and extrapolating it as some deeper insight, if I'm honest. That impulse is strong. It almost feels natural; like falling toward a black hole. Maybe it's some kind of weird psychological rite of passage when you start using a coding agent.
To be clear: sharing your personal experience is great; that's how communities learn. What I'd push back on is packaging those experiences as prescriptive guidance when none of us have been doing this long enough to know what actually works at scale.
Swimming against the current, I think it's important to stay critical, humble, and open-minded right now. Just try things... experiment. Make useful shit with these tools, but hold onto your values and principles. It's more like going to the gym than engineering: show up consistently, do a bit more than yesterday, don't compare your progress to someone else's highlight reel, and don't expect a training program that works forever (or even for multiple months). The mental model you build today might be obsolete even tomorrow...
November is Florida Veterans Appreciation Month, a time to honor and celebrate the achievements and sacrifices of our Veterans across the state. Letโs recognize their dedication, resilience, and contributions. Join us in saying thank you to Floridaโs heroes! #FDVA#FLVets๐บ๐ธ
@kurtbuhler do you have a post on power bi project documentation? At the end of a reporting project it's time for a knowledge transfer, any templates you recommend? i.e. documenting sources, ETL, semantic model, reports, governance, etc...
NEW blog analysis by @dhinchcliffe: @Microsoft's #AI and Copilot Announcements for the #Digital Workplace https://t.co/tmIF3fsBfd Announcements at Microsoft Build 2023 centered around the adoption of #AI across Microsoft's various products.
@KratosBi "Since OneLake is powered by a capacity, the new Direct Lake capability of Power BI is available only when used with Fabric or Power BI Premium capacities."
https://t.co/GPWfaExshm
@KratosBi This is still great news, Direct Lake is available with the lowest level entry F2 capacity, which I assume is what Amir referred to as the $180/m option, right?