Dr Nic Williams

Today's Xero TechTalks session with Doug Guthrie from Braintrust was incredibly insightful. We explored the crucial topic of effective LLM evaluation, covering everything from defining tasks and "good" outputs to implementing data-driven improvements. Doug shared practical workflows, from early validation in the Playground to production monitoring, with a key emphasis on the often under-utilised practice of error analysis. Thank you Doug Guthrie, for a fantastic session! #TechTalks #AI #LLM

7

What about the continuous review cycle, recurring 1:1s, and the ability to “move people out” at any time in WA state? The single reason for a bulk reduction in force is cost-cutting, and they’re using performance as an excuse. Talk about perfromative… I can’t believe people buy this. Seattle-based Zillow cuts 200 jobs https://lnkd.in/d7G32wdQ

9

2 Comments

DDD Melbourne speaker spotlight: Callum Whyte Failure will happen. In software, that’s not a risk—it’s a guarantee. Code breaks, servers crash, power goes out – yet building software that is designed to fail gracefully isn't always front-of-mind. Resilient systems don’t avoid failure… they’re designed to fail gracefully. Callum Whyte will walk us through how to design and build real-world systems that keep working when things go wrong. Expect a practical, engineering-first deep dive into: – Handling failure at every layer – Redundancy, circuit breakers & automated failover – Architectural tradeoffs: latency, consistency & degraded modes – Building reliability into team processes and mindset – Using health checks, OpenTelemetry & stress testing tools to stay ahead of issues This session is packed with tools, patterns, and techniques you can bring back to your team immediately. Learn more about Callum's session: https://lnkd.in/gCmg-8sU Grab your tickets to DDD Melbourne now before they sell out! https://lnkd.in/gC4yyEbF

11

🔥 New article: The Hot Potato Programs Ever been handed a project so chaotic you wondered if you were on a hidden camera show? I’ve been there, more than once. Recently, I led the world’s largest Copilot Studio implementation, supporting over 15 million customers at Australia’s biggest financial institution. Coordinating engineering, sales, product, consulting, vendors, and specialist teams felt like juggling lit dynamite. Here’s the truth: no shiny framework rescues a failing program. Clarity, sequencing, and trust do. In this article, I break down how to: ✅ Depersonalize the mess without losing urgency ✅ Quantify what’s broken in plain language ✅ Stabilize before you build ✅ Reset vendor relationships ✅ Show progress early: even if it’s small If you’ve ever inherited a “hot potato,” this one’s for you. 👉 [𝗥𝗲𝗮𝗱 𝘁𝗵𝗲 𝗳𝘂𝗹𝗹 𝗲𝘀𝘀𝗮𝘆]: https://lnkd.in/gB8jJbXc ♻️ Repost to help others learn these valuable lessons about program management. ➕ Follow RAJESH MATHUR for more.

34

16 Comments

Bun’s trajectory has been toward a cloud‑native, self‑contained runtime: single‑file executables, builtin S3/SQL/streaming clients, fast startup, strong Node compatibility. That matters when agents like Claude Code need to install, run, and test code across heterogeneous environments with predictable performance. Net effect: a faster, predictable runtime and tooling layer built around AI agents, with long‑term stability and resourcing while remaining open‑source & MIT‑licensed.

4

"Building NDIS software must be hard." Yes, writing the actual code is hard but it's not the hardest part. The biggest struggle is understanding all of the details and edge cases when there is so much variation is in how businesses actually operate in the real world. One day you’re talking to a 5-person team using spreadsheets and doing payroll manually. The next, a 300-person org has 6 layers and automated payroll. Those businesses operate extremely differently. One key to building a successful software platform is actually talking to your customers (and potential customers) about the problems they are facing, not their current solutions. I don't mean automated emails and surveys and questionnaires; while they are important and they do help. I literally mean picking up the phone or getting on a video call and asking, "Walk me through the struggles of your payroll" and trying to understand every single detail. Frank, Jacques and the team at Kinso AI is doing this so well at the moment and they don't even have a product launched. Linear does an amazing job at this with their customer Slack channel. Ken and the team at Clipflow are asking all the right questions to those struggling with content management. I've recently seen Jared Palmer do this for GitHub on X for engineers. When customers use generic words like: • "It's painful." • "It's clunky." • "It takes a long time." We need to dive further into that. What is clunky? What part of the process is taking the time? Getting the size of your sample pool is challenging. You need a pool that is big enough to give you the best understanding of the problem, but you don't want a pool that is so large that it just generates too much noise. I have spoken to literally hundreds, if not over a thousand, NDIS business providers... Yet every single week I still find a customer or potential customer that is doing things differently - not good or bad... just different. And that's when I pull this face in the photo ⬇️ 😯 Our job is to find the common ground between all of the problems in the NDIS and solve them for businesses in a way that doesn't suck. And that's exactly what we're doing.

16

2 Comments