Catch-up: what we learn from Sonos and OpenTofu, how I use AI at work, Bluesky is getting pretty good.

Agentic AI

I’ve been spending a lot of time thinking about the Tanzu point of view on AI and the stack we have (Spring AI and the VMware Tanzu AI Solutions). There’s a lot mysticism around agentic AI, but when you reduce it down to an API, you can simplify it. Spend less time thinking about fully autonomous AI that, like, writes its own code, and more just focusing on putting AI into enterprise apps.

What makes it enterprise AI is that you architect and manage it to be used by hundreds and thousands of apps and people in regulated organizations. These are not tech companies, and I mean that in a good way. They’re the organizations that keep lives tidy and keep the lights on. I like a clean, well lit place. This means you want to build platforms, pay attention to cost, accuracy, and as many -illity’s as you can afford.

There are some core patterns you need for any enterprise app, and just making sure you can handle them in your AI-enabled apps will be enough work for the next five years, if not more. There’s, like, eight or ten basic functions to agentic AI. And then you need to get access to models (local or in public cloud): you need an AI broker. Pick a framework and a broker start using it.

In my world, large organizations, Java dominates and, thus, Spring is extremely popular, so I’d recommend that. It’ll change and adapt very quickly to whatever comes along, and then you can protect yourself from shifts in the future: Spring will be a safe boat in the weird waves of enterprise AI.

But, what is agentic AI?

Here’s three, recent definition:

1. "At its core, the concept of an agent is fairly simple. An agent is defined by the environment it operates in and the set of tools it has access to. In an AI-powered agent, the AI model is the brain that leverages its tools and feedback from the environment to plan how best to accomplish a task. Access to tools makes a model vastly more capable, so the agentic pattern is inevitable." –Chip Huyen, author of AI Engineering, January 7th, 2025.

2. "'[A]gents' has become a loosely defined term in the post-ChatGPT era, often referring to LLMs that are tasked with outputting actions (tool calls) and that run in an autonomous setting…. [T]hey require state management (retaining the message/event history, storing long-term memories, executing multiple LLM calls in an agentic loop) and tool execution (safely executing an action output by an LLM and returning the result)." –Letta blog, "The AI agents stack," November 14th, 2025.

3. "[A]gentic AI, which uses sophisticated reasoning and iterative planning to autonomously solve complex, multi-step problems…. Agentic AI systems ingest vast amounts of data from multiple data sources and third-party applications to independently analyze challenges, develop strategies and execute tasks." Erik Pounds, NVIDIA blog, October 22, 2024.

This means looking trendy agentic AI. What an award thing to have to type each time: do you craft your sentences so you write “AI agent,” or just dive into the absurdity of “agentic AI”? Also, as you look at what agentic AI is evolving to, it’s quick becoming “all of AI.” Just doing one-shot generative AI prompting is not that interesting or useful, really. At the very least, you need to store context and memory, and then have access to tools and functions…and then you’re off into agentic AI.

While the definition of an AI agent will continue to evolve, here’s some key aspects:

1. The agent is/can be autonomous.

2. The agent uses data, context/environment, memory.

3. The agent plans and takes actions.

Agentic AI theory is trying to get more from generative AI, to get it to "do things."

Always focus on the work actually being done. Ask “what is the app and what will it do?”

Wastebook

• “Face computer” v. “skinny computer.” Om on Apple.

• Mutesignaling.

• “Exhausted Majority.” Charts.

• “Does content exposure to this get me closer to the person I want to be?” RotL #567.

• “John Fetterman’s Shorts.” Casual drsesscode.

• “A preference cascade” Here.

• “There’s a melee of dwarves on the floor to each side of you.” Dust up in Citadel Akbar.

• They don’t mean it, or even know they’re doing it, they’re just a casual asshole.

• For ever brain that is drained, there are three brains waiting. Any brain that wants to drain should feel free to: keep moving and get out of the way.

Relative to your interests

• Building Effective Agents with Spring AI - rapid development going on in the Java/Spring community with AI.

• Best Prompt for Academic Papers Summary/Analysis - “Explain this paper to me using the Feynman technique, considering yourself as the author”

• Share of teens using ChatGPT for schoolwork doubled from 2023 to 2024 - It’s great for all your education needs. // Worth considering in any discussion of generative AI for education is: compared to what? In my experience, there are not enough nor enough access to meat-teachers. The AI is always there. This mans there is more access, when students need it. That is not bad.

• Common pitfalls when building generative AI applications - Good list, and not just the general purpose pitfalls.

• Protecting your mental health during a clown president’s second term

• Trump signs order setting up DOGE with a focus on government tech - ‘Now, USDS is set to work on a “Software Modernization Initiative,” per the executive order, “to improve the quality and efficiency of government-wide software, network infrastructure, and information technology (IT) systems. Among other things, the USDS Administrator shall work with Agency Heads to promote inter-operability between agency networks and systems, ensure data integrity, and facilitate responsible data collection and synchronization,” the executive order read.’

• 2025–01–21 - “I do not have a solution for any of this stuff, and I despair as much as the next sane human being, but one thing has become clearer and clearer since 2016; social media, like Monty Python’s Camelot, is a silly place.”

• “Be careful what you wish for, you might get it!” - Any weapon you make can be turned against you.

• Gartner sees 10% IT spending jump in 2025, but don’t get too excited

• How pointy shoes created a moral panic in medieval London - It’s good to remember that hundreds of years later, and prohibition against fashion makes the probities look like idiots with too much time on their hands. // Also, Dutch businessmen don’t seem to have stopped wearing them.

• New D&D Book Releases in 2025: Monster Manual, Anthologies, and More

• - Researchers are questioning if ADHD should be seen as a disorder - It’s better to think of most psychological friction (“problems,” you don’t bristle at the term) in reverse: what is the system doing wrong that’s causing people to act poorly. // This is especially true in the workplace where profit doesn’t give a fuck about curing and would rather do whatever it takes (within the law) for the desired business outcome.

• Abscission - Taking care of your mental state when you quit a job.

Conferences

Events I’ll either be speaking at or just attending.

cfgmgmtcamp, Ghent, Feb 3rd to 5th, speaking. SREday London, March 27th to 28th, speaking. Monki Gras, London, March 27th to 28th, speaking. CF Day US, Palo Alto, CA, May 14th. NDC Oslo, May 21st to 23rd, speaking. KubeCon EU, April 1st to 4th, London.

Discounts: 10% off SREDay London with the code LDN10.

Logoff

Fintan started a daily links log. I love daily links logs! I get a lot of links and enjoy two: Richard Seroter’s and Assaf Arkin’s, as noted in my colophon. It’ll be cool to see what Fintan catalogs, if you like my weird pile of links, you’ll like the previous two and Fintan’s, like and subscribe!

//

Several of my talks were accepted, so you’ll see some more conferences above.

//

I saw someone doing that “ICYMI” thing at the top, so I thought I’d put a “Catch-up” line in there. Beats me, I just want them clicks.