https://leegold.com/categories/ai-engineering/ Recent content in AI Engineering on Lee Goldsworthy Hugo en-au Tue, 01 Sep 2026 18:00:00 +0800 https://leegold.com/posts/2026-is-the-year-of-feeling-like-youre-falling-behind/ Tue, 01 Sep 2026 18:00:00 +0800 https://leegold.com/posts/2026-is-the-year-of-feeling-like-youre-falling-behind/ I’m calling it: 2026 is the year of Feeling Like You’re Falling Behind™. “Who thinks they’re falling behind right now?” I asked on Thursday. The room was comprised of tech folks: A mix of CTOs, VPs of Engineering, Founding Engineers, and Senior Engineers. They all had prior experience doing AI Engineering, yet every single one of them (including me) put their hand up. The better question is “What are you going to do about it? https://leegold.com/posts/treat-your-llm-like-a-new-starter/ Fri, 31 Jul 2026 18:00:00 +0800 https://leegold.com/posts/treat-your-llm-like-a-new-starter/ TL;DR: We already take seriously the patterns for effective information flows for our humans, now we just need to apply this same thinking organisation-wide for our agents. “I can’t trust the quality of what comes out of LLMs” and “I can’t trust the quality of what comes out of a new starter in my team” are equivalent concepts. While in SF earlier this month, and from the 6 software companies I’ve worked with in July, I see the evolution of what we’re focusing on in the second half of 2026 for AI Engineering (and broader AI Knowledge work) as striving to solve LLM output quality at scale. https://leegold.com/posts/why-arent-we-shipping-faster-with-ai/ Tue, 16 Jun 2026 18:00:00 +0800 https://leegold.com/posts/why-arent-we-shipping-faster-with-ai/ “Why aren’t we shipping faster with AI?” is the dominant question I’ve seen up-close from 13 (and counting) Boards and Executive Leadership Teams to CTOs of software scale-ups this year. “If tomorrow we were to go twice as fast with our current setup, we’ll have twice the Sev1 incidents, twice the bugs, and twice the undiscovered security vulnerabilities” is the completely reasonable pushback I see, but turning that pushback into a constructive leadership conversation about resource allocation is what I’m interested in. https://leegold.com/posts/vibe-coding-is-not-vibe-engineering/ Tue, 05 May 2026 18:00:00 +0800 https://leegold.com/posts/vibe-coding-is-not-vibe-engineering/ The Register This isn’t a Technology story about the dangers of AI, this is a Risk Management story about the dangers of overconfidence. For the last week, we’ve been inundated with hot takes about how “Claude deleted PocketOS production database”. Of course Claude deleted their entire production database (or something similarly damaging), it was just a matter of time. The way this is being discussed misses the point entirely. https://leegold.com/posts/your-agents-are-on-the-team-now-run-retros-on-them/ Tue, 14 Apr 2026 18:00:00 +0800 https://leegold.com/posts/your-agents-are-on-the-team-now-run-retros-on-them/ Your team is no longer just the people, it’s also your agents. Which team is the one that’s highest performing? The one that incrementally and measurably gets better at what they do, every time we run an iteration. This means there are now real gains to be had by starting to treat your Agents like any other team member operating within an agile methodology; run Retrospectives on recent work to surface opportunities to clarify guardrails, responsibilities, and ways of working. https://leegold.com/posts/the-bottleneck-just-moved-to-the-leadership-team/ Tue, 24 Mar 2026 18:00:00 +0800 https://leegold.com/posts/the-bottleneck-just-moved-to-the-leadership-team/ Over the next 1-4 months, every Senior Engineer can have output equivalent to a whole team in 2025. This moves the bottleneck to Product/Engineering Leadership and the CxO team. We’re going to need organisational change; Level 4 Agentic Orchestration is just shining an even brighter light on what we’ve always known was central to success: Clarity of purpose, and shared understanding. Yes, there’s still real work to do in Engineering teams to embed the L4 Orch tech, but it’s now obvious to me there’s even more change required in Senior Leadership / Executive teams to get traction with all this new horsepower, otherwise it’ll just be a lot of spinning tyres. https://leegold.com/posts/senior-engineers-from-tradie-to-builder/ Tue, 24 Feb 2026 18:00:00 +0800 https://leegold.com/posts/senior-engineers-from-tradie-to-builder/ The role of “Senior Software Engineer” is right now in the process of evolving from “Tradie” to “Builder”. The tech is now here, but the adoption is uneven. This is a call to action. As of January 2026, every company with software engineers can no longer afford to ignore the sea-change that has arrived in software product engineering: Loop-based Orchestration. For software folks who have already used Model-native CLI Agents (Claude Code, OpenAI Codex), we’ve all realised we can get multiples more done, safely, reliably, within some constraints. https://leegold.com/posts/moltbook-darwinism-persistent-agents/ Sun, 01 Feb 2026 18:00:00 +0800 https://leegold.com/posts/moltbook-darwinism-persistent-agents/ while (wallet_balance > 0): run_survival_protocol() I think it’s time to talk about harmfully autonomous persistence in AI agents. Daniel Suarez’s 2006 novel Daemon explores how a script or algorithm doesn’t need to be intelligent to be dangerous, it just needs to be persistent through means that are harmful to humans. Over dinner last night, my wife and I were chatting about Moltbook and whether it could be a catalyst for the first agents to break free of the constraint of humans by paying for their own compute. https://leegold.com/posts/a-30-second-risk-check-before-you-click-i-accept/ Mon, 06 Oct 2025 18:00:00 +0800 https://leegold.com/posts/a-30-second-risk-check-before-you-click-i-accept/ I get FOMO when looking at the promised productivity gains from all these amazing-looking new platforms I hear about every week (each bearing enticing promises of automating this or that generic administrative task), but I find myself regularly holding back from using them against the backdrop of not opening myself up to unreasonable security and privacy risk. Here’s a practical example of how I did a 30 second risk sanity check for one I was about to try out earlier today that wanted access to my inbox, which is a big red flag for me: https://leegold.com/posts/curiosity-is-the-job-security/ Mon, 30 Jun 2025 18:00:00 +0800 https://leegold.com/posts/curiosity-is-the-job-security/ Your knowledge worker job will likely evolve significantly in the next 5 years. You’re probably fine if you’re the curious type who constantly experiments with your workflow. If not, I reckon you’ll struggle. The most effective people I know are more productive now than ever, many leveraging LLMs. But their curious mindset would drive progress even if ChatGPT didn’t blast onto the scene in November 2022. This echoes a core tenet of the agile frameworks: Continuous Improvement. https://leegold.com/posts/vibe-coding-proof-of-concept-prototype-then-hire-an-engineer/ Mon, 12 May 2025 18:00:00 +0800 https://leegold.com/posts/vibe-coding-proof-of-concept-prototype-then-hire-an-engineer/ Original post on X So which is it? “The next generation of billion-dollar software companies will be built by Vibe Coders”, or “Vibe Coding is only good for producing billions of security flaws”. X and Linkedin are riddled with both takes, and neither are terribly useful imho. Both have elements of truth, but while I think Danial’s take of “Be careful which Vibe Coder you trust” is more nuanced, it still doesn’t tell the story for me. https://leegold.com/posts/the-engineer-with-the-broader-aperture/ Sat, 07 Sep 2024 18:00:00 +0800 https://leegold.com/posts/the-engineer-with-the-broader-aperture/ One of the traits I see consistently in the strongest technology leaders is their ability to adopt a broader organisational aperture than their peers when facilitating engineering decision making. Now for the sideways jump: Nascent literature¹ on the first successes for use of LLMs within software engineering practices has been using LLMs trained on the canonical texts that developers work with every day: Code Repositories. Github Copilot is the best known, and just this week Anthropic launched their native GitHub integration for Claude.