AI, Carefully Applied | ✉️ #84
Hey! 👋
It’s customary at the end of the year to take stock of things. What worked out as planned, and where the plans turned out to be a bit more ambitious than reality allowed. I, however, have an interesting observation about how different large corporations have been summing up their year internally. No names, of course.
And in almost every case, the main focus was on whether they managed to ride the AI wave and integrate it into their processes. What’s curious is that no one seems to have any meaningful metrics for evaluating this integration. In theory, it should be something like: X number of processes handed over to AI automation, or these specific tasks are now fully handled by generative technologies. None of that exists. The assessment is basically: “Yeah, we tried it, here’s what we got.”
The biggest success story I’ve seen so far probably belongs to compliance departments. Every large enterprise has rolled out a few internal trainings on how to use AI and how to spot red flags. About 60–80% of those trainings revolve around a single idea: AI will confidently bullshit you, so be careful before forwarding its output to your colleagues. By the way, we run these kinds of trainings too — feel free to reach out.
Interestingly, this overlaps quite neatly with my own observations. In the business world this year, AI has been used very differently from how it’s portrayed in posts on X/Twitter. Sure, there are viral cases where yet another fashion brand is dropping another piece of AI slop on Instagram, or those Coca-Cola and McDonald’s ads that triggered waves of backlash. But real-world usage (beyond writing reports and emails) mostly boils down to storyboarding and moodboarding that no one outside the corp sees. And every such AI use case comes wrapped in multiple disclaimers: this image, this part, this fragment was generated with AI for demonstration purposes only and does not represent [insert final product here]. This kind of usage has quietly spread across almost all levels inside companies and in fact speeds up business processes. So if you think AI is mostly used by developers and marketers — that’s not the case. And for me, that’s the main takeaway of the year.
I talk about another example of how people with little to no connection to IT are using AI in the latest episode of our DevOps Accents podcast, where we sum up the year. Don’t miss it — it turned out to be a fun one.
And with that, see you next year.
Happy New Year! 🎉!
What We've Shared
DevOps Accents, episode 67: 2025 Year In Review. In this episode of DevOps Accents, Leo, Pablo and Kirill look back at their own predictions for 2025 they made in 2024 and compare it to what actually happened:
Terraform Won. Kubernetes Won. Your Tech Choices Are Over. A video segment from the previous episode of DevOps Accents for you!
Unlimited image generation with Nano Banana Pro and custom Claude Code Skill: Nano Banana Pro might be the best image gen model right now — but the real trick is how Kirill wired it into Claude Skills. 100+ icon iterations, 4K control, self-critiquing generations, and sane context handling. $45 well spent.
What We've Discovered
Cloudflare outage on December 5, 2025: Cloudflare’s post-mortem — what went wrong, and how it affected ~28% of HTTP traffic. We plan to discuss this in detail in the next podcast episode, but for now read these practical takeaways for architects who don’t want their site to be single-point dependent on someone else’s edge.
EU publishes first draft Code of Practice on transparency for AI-generated content: If you produce, detect or ship synthetic media this is your essential reading.
The 85th mkdev dispatch will arrive on Friday, January 16th. See you next time!