Week 3, 2025 - New Horizons
It was a busy week. Indeed, it was so busy that I overplanned my Friday and didn’t have enough energy left to finish and send out this newsletter. Usually, I don’t like to work on Saturdays, so this is an exception.
This week I left behind thousands of followers on X and moved to Bluesky. There are increasingly more arguments to do this move, most of which I agree with. But the strongest driving force was that simply put, Twitter is the past. I’m not sure that Bluesky is the future, the developments around the Fediverse are super exciting, but for now, it seems that the content and the experience is more mature here. You can find me at @peterszasz.bsky.social.
Last week I wrote about focusing more on my professional offering. A first step is here, even if it’s just a self-hosted placeholder page for now: Leadtime.tech.
I admit, I am obsessed with names with layered meanings. Getting inspired from one of the key DORA metrics, while capturing the moment when a techie realizes they have ambitions to consider management — this was a good enough choice to call my business and solve one of the hardest problems in computer science: naming things.
📋 What I learned this week
Most of my 2024 was spent with ad-hoc and instinctive progress towards a consulting goal of mine, with a lot of exciting projects giving me more clarity on what I want to do. However, one of the things I want to change in 2025 is to be more strategic and intentional. I need a good framework to discover what I want to do and how to do it, and capture these in strategic documents.
I started with V2MOM from Salesforce. This framework seems to nail the balance between depth and complexity. I experimented with AI assistance to guide me through the process, using the Projects feature of Claude. I filled its context with two kinds of documents:
- What I want to do: descriptions of V2MOM, examples, and similar materials.
- Who I am: my CV listing my experience and achievements; the CliftonStrengths report I received after doing the assessment a while ago, showing areas I’m good at, and things I should navigate around carefully; and some sample articles I wrote that capture how I think about management.
Loaded with this context, I gave the role of Claude to be an expert coach who specializes in helping seasoned engineering leaders who start a consulting and coaching business set up their strategy with the V2MOM framework.
In some sense, this experiment was a hit-and-miss. Letting my guard down and just answering passively the questions often led the discussion to inproductive, tangential rabbit holes. In a long session, we only talked about personal principles, until I got frustrated and reminded Claude of the ultimate goal of these sessions. However, a minor adjustment of the initial role-defining prompt, which forced the model to challenge the things they heard from me, holding a tighter control on the process, helped me discover surprising obstacles and areas I need to pay better attention.
The lesson for me is this: using large language models for strategy-setting assistance has many pain points, and success depends on iterative initial prompt refinements as much as the discussions themselves. It doesn't replace an expert at all. Yet, it’s still a million times better than struggling with this alone, without help or feedback.
There were also a lot of learnings about the prompts themselves that I might publish one day as a follow-up. It’s still a developing process for me, though I had a lot of initial success in managing the length of conversations and maintaining context through different sessions.
Continuing last week’s train of thought, I settled with the Session App to manage my Pomodoro Sessions. I’m not done summarizing the experiment, but so far I’m happy with what I see. I love the extendability of the app, for example. I like to store all my calendar-related data, including these sessions, in Google Calendar. The Session App lacks this feature, but thanks to their support of Apple Shortcuts, I could throw this simple synchronization shortcut together easily:
All it does is when triggered by the end of a session (which is automatic), it records the session's data in a local calendar that’s set up using Google - ready for me to process and get insights from, eventually.
🎯 What I want to try next week
The week should mostly be about Leadtime.tech strategy finalization. I have a small pet project in the quantified self area, if I have some time left, I’d like to reach a milestone where I can write about it.
🤔 Articles that made me think
Laziness, Impatience, Hubris, and AI
Arctortect is describing a challenge with relying on LLMs in a candid, self-reflective way, sharing the struggles of gaining short-term productivity feeling at the expense of losing long-term craftsmanship skills, almost like a Faustian deal.
I also think the solution lies in the efficient use of the tool: to keep strong control of the creative process and allow AI to only do the smaller, straightforward tasks, in micro-steps, interactively and iteratively. But the temptation to throw caution to the wind is strong indeed, because the results are quick and shiny, giving an image of effortless productivity – until the tool’s limits are reached and the illusion falls apart. (The title also gave a nice trip down memory lane - I used to have a t-shirt with an obfuscated Perl program that typed it out back in the early 2000s, when I believed in the false image of a superhero programmer.)
On the recent GitHub outage
This Bluesky post about an engineering team taking a break from work because GitHub was down made me think that there’s something fundamentally wrong if you have a single point of failure at the setup of your git provider. (A distributed system by default!) It’s not like an AWS downtime when half the internet is out for your customers anyway - it’s impacting only you, in the sense that you can’t ship code to your users. It's either an oversight or a calculated risk – but in both cases, the responsibility is yours. (Also, the cost of adding a little robustness in your git setup is much lower than going multi-cloud.)
At the minimum, two criteria should be satisfied in most product development companies:
- engineering teams should be able to continue interacting with the codebase;
- processes and documentation should be in place for emergency deploys.
S3-compatible backups for PikaPods
More of an announcement, that thought-provoking article, but I was really happy to see this happen. PikaPods, the hosting provider for this blog implemented s3-compatible backups, one of the most requested features. All I had to do was create a new B2 bucket in my Backblaze account, grab its API key, and fill it into my pod’s config. Under the hood, they use Restic, the same tool I wrote about handling my backups. Great news. I love how responsive this small provider is and this user-centricity and their competitive prices make me a happy customer.
(Product organizations, take note: they released this feature in an early stage, without a key backup functionality: automatism. Currently, all I can do as a customer is to set up and trigger manual backups. They decided to launch now and observe how it's used, to be able to adapt before finalizing the feature. This allows them to be much more flexible and respond to real-world user feedback, without the burden of maintaining a more complex feature list. Love it.)
🎬 Something cool: David Ehrlich’s 25 Best Films of 2024 Video Countdown
Part movie top list, part video clip, part movie nerd easter egg hunt, part short film: David’s end-of-year best movie video countdown is a piece of art in itself, and worth a watch even if you don’t know half the films on the list. What a masterpiece of editing!
That’s it for today, stop and look at the blue sky this weekend,
Péter