Kurt Buhler 🇨🇦

We're currently discussing the best way to share and explain these observations and practices in a high-quality -- but also responsible -- way, and making sure that they're not specific to our scenarios and opinions.

TLDR agents are cool, powerful, fun... but they're loaded guns.

It also necessitates a certain process structure and workflow maturity, otherwise it's just treadmilling tokens. This works best when it's tailored to your scenarios and not generalized across use-cases and tasks (esp. greenfield vs brownfield agents).

That's not that it's not valuable or useful... Rather, to get it to "work" you need to invest tremendous time and effort in building good context, tools, and prompts (even optimizing with model selection for different tasks)... enough for it to "know" but not too much that it "rots".

However, the high does not come back.

Applied to projects (and not demos) it can quickly, catastrophically fall apart w the smallest things... not even in big, clear ways but subtle, nuanced mistakes that you might not notice until much later.

Like assembling IKEA furniture without instructions.

The first time one generates something that works, it's like this big high -- "oh my god, it worked... what else is possible?... AI can do this now!" A few Contoso / AdventureWorks demos, and suddenly the lens is too dirty to see reality.

The reason I do not make content on agentic coding tools (despite using them daily) is because the moment folks see agents manipulating files and doing tasks that *look* correct, nuance and skepticism go out the window.

It's like demo heroin. Right now, it just feels really fucking irresponsible.

it does for sure

www.linkedin.com/in/julien-cd/

Give Julien a follow on LinkedIn. Really nice, genuine person with some original and interesting ideas applied to Power BI.

I'm skeptical of the actual representativeness of this chart from FT about how american personality traits are evolving over time.

Nonetheless... concerning trends.

said it many times and ill say it again -- running ttRPG games for the last 10+ years has been the single most beneficial thing for my consulting career. The soft skills development has been incredibly valuable.

You can't use Sonnet 4?

hahaha amazing

I find it highly amusing that two things seem to be true at the same time:

"You just talk to AI in natural language! It just works!"
and...
"Prompting is a skill! Of course you're not getting the right results, you have to get better at prompting properly!"

Answer in the middle.

AI influencers and techbros sound more like megachurch pastors every day.

When momtana met dadkota.

I love to live in the future ....

This is - sadly - what I'm thinking. Also distracts from the fact that they were building up to GPT-5 being some leap or even AGI claim and it nothing of the sort was demonstrated.

Pretty soon consultants are gonna start charging different rates for (with thinking) and (without thinking), too.

Unironically the worst charts I've seen off of Fox News in America. I'm still trying to figure out if it's some kind of joke, or intentional weird viral marketing strategy.

The graphs from the GPT5 announce are so, so funny.

Next article by Johnny Winter about using Tabular Editor with Databricks! tabulareditor.com/blog/tabular...

Part 2: tabulareditor.com/blog/tabular...
Part 1: tabulareditor.com/blog/tabular...

#TabularEditor #Databricks

The Ugly Truth About AI Coding Tools
medium.com/@amareshadak...

The question for AI users is where can it work well enough to be useful. Domains that are logical and factual remain a poor match to the statistical algorithms of LLM.

survey.stackoverflow.co/2025/

Over-discussed: "use AI to vibe code an app/idea"

Under-discussed: "ship a product that customers love [because it solves a persistent pain point, is delightful to use, and just works]"

Without the latter outcome in mind, you're mostly wasting time with the former...

The real glory of this new "age of AI" is you can quite quickly make small, private tools for you by you to enable you and your team/org/client. Open sourcing shit is a commitment that you need time, resources, and energy for... and sometimes its not ethical to just kabouter plop something out there

My brother in christ you can literally yeet your tenant into the void with this I'd love to give it to you because that would be hilarious but no. lol no.

I have various MCP servers which together have over 100 r/w tools for Power BI/Fabric. In my SQLBI video I show this briefly; since then we've had quite a few people asking -- rather, demanding -- access to these servers. Some quite aggressively.

"Gimme the loaded gun so I can shoot myself"

only if you factor in the 1 min memory until you enter non euclidean nightmare space haha

this but models and reports. 😎👉👉 consulting bodyshopping 2026

Generative AI tools are the blood diamonds of the tech industry. The technology is quite impressive and valuable but the ethics of how it’s made is questionable AF.