Micros
Short micro blog pieces that might lead to something bigger...
The Reserve Bank of Australia's October 2026 Financial Stability Review points to incident risks from “complex supply chains and operational dependencies” from entities beyond the regulated financial sector.
“Incidents can arise at material service providers and propagate through complex supply chains and operational dependencies, requiring authorities to consider a broader set of entities beyond the regulated financial sector. ”
Is this an opportunity for software engineers to add value by proactively monitoring dependencies?
I think so, and wrote about dependency monitoring here: Dependency monitoring
(No doubt others have said it better than me. For example: Alex O'Callaghan)
Some AI hype may be overblown. After all, in the 19th century it was predicted that railroads would span continents. https://en.wikipedia.org/wiki/Cosmopolitan_Railway
Learning techniques ranked by effectiveness (according to research):
- High: Distributed practice
- High: Practice testing
- Moderate: Elaborative interrogation
- Moderate: Imagery
- Moderate: Self explanation
- Low: Mnemonics
- Low: Re-reading
- Low: Interleaved practice
- Low: Underlining
- Low: Summarization
(From meta-analysis of 242 studies)
A Meta-Analysis of Ten Learning Techniques • Frontiers in Education
How I've been using AI for broad research:
- Find passionate Reddit community around a topic and download its FAQ and Intro
(Simply append
.jsonto the URL and download the file) - Find peer reviewed journal articles on the topic and download them as PDFs
- Create a Project in Claude and drag-drop the downloaded files to Context
- Prompt Claude with specific questions on the topic (and ask for sources)
Second thoughts on AI generated specs... maybe they are not far off from being effective. They key is humans thinking more probabilistically. Specs don't have to be 100% correct and they don't have to be correct 100% of the time. They just have to be correct enough to improve some business metric(s). Maybe we are entering an era of probabilistic business operation. In a way, maybe it was already this way. We humans like to think we know the exact answer, but in reality, everything is a bit fuzzy.
Are strong, deterministic, highly specified guardrails (e.g. context, tests) really required to keep AI on track? If so, then aren't AI engineers applying software engineering to guardrails?
It seems that AI engineering is basically software engineering applied to building AI guardrails.
- Similar to how cloud engineering = software engineering applied to cloud infrastructure.
- Similar to how test engineering = software engineering applied to test infrastructure.
See the pattern?
Software engineering fundamentals + specific knowledge in a sub-field.
Fundamentals don't disappear. Rather, specialisations are added to them.
Why startups fail – the 8 patterns of failure (according to Eisenmann).
- Catch-22: Founder cannot pursue opportunity without resources, but cannot attract resources until they've pursued the opportunity.
- Good Idea, Bad Bedfellows: Startup with a promising opportunity falters due to deficiencies and dysfunction among key players (e.g. founders, team members, investors, strategic partners).
- False Starts: Startup rushes to launch its first product before conducting enough customer research—only to find that the opportunities they’ve identified are rife with problems.
- False Positives: Startup falls victim to premature scaling; expands too rapidly, without adequate resources to operate successfully at a higher scale.
- Out of the Frying Pan: Late-stage missteps due to imbalance of speed and scope.
- Speed Trap: By expanding at an unsustainable pace, late-stage startups fall prey to a speed trap.
- Help Wanted: Startup sustains product-market fit as it grows but cannot mobilize the resources needed to continue expanding.
- Moonshots and Miracles: Startup pursuing bold innovation faces many big challenges and a shortfall with any one of them will be sufficient to kill the venture. Consequently, the startup needs a cascade of miracles to succeed.
From the book: Why Startups Fail • Tom EISENMANN.
Installing Obsidian, adding my pre-existing notes folder and everything just working is a great #ux! An examplar for software interoperability.
Brilliant talk! The “debt” metaphor in “tech debt” is problematic. Code quality is best seen as a continuous process, which needs to be supported by frequent delivery, appropriate timeframes, continuous learning, transparency and advocacy.
Thought: CSS inheritance seems to cause similar maintainability issues to OO inheritance.
Good systems tend to have a lot of redundancy (“fall-backs”), and thus, have some inefficiency. Designers often seem to understand this better than engineers.