Web Directions Year round learning for product, design and engineering professionals

There’s less than two months until AI Engineer Sydney on 7 and 8 December, and our programme is really coming together.

The strength of the proposals made selecting presentations extremely challenging, but the set of talks that emerged is incredibly strong. Together they paint a picture of what practitioners, both engineers and engineering leaders, are actually wrestling with as AI moves from experiment to production.

So here’s a sense of the programme as a series of themes and interrelated conversations, rather than an isolated collection of talks. These are some of the themes we’ve seen emerge, with some of the key talks for each. In future posts we’ll drill down into them more specifically.

Sovereignty has become an engineering question

Sovereign AI is part of a much broader national conversation, and it has reached the engineering team. What sovereignty actually means is very much in the eye of the beholder, though. Australia is spending billions on sovereign AI infrastructure. But as CommBank’s Leo Borges puts it, sovereignty isn’t about where the GPUs sit. It’s about who can switch the intelligence off. Leo connects national AI strategy to production architecture, and what an open-weight fallback looks like in practice. Mark Pesce asks what happens when “good enough” models run on your own machines and the meter disappears.

To put this into practice on the workshop day, Joseph Gulay will show us how to move an agentic coding workflow onto open-weight models running on Australian infrastructure.

The economics of agents are catching up with everyone

Token bills are now a design constraint. Google’s Tanya Dixit shares the architecture patterns behind cutting agent token costs by around 70%. Hello Clever deleted most of its agents and cut cost per task by about two-thirds.

At Canva, Fawaz Ahmad argues that token usage is the new lines of code, and just as useless as a measure of whether AI is actually paying off. And Google’s Daniel Nadasi looks to economic history for what this transition means for software work, and for the emerging human and agent workforce.

Trust is engineered, not assumed

Suncorp brought on a dedicated AI red-teamer before APRA and ASIC told organisations to lift their AI security game. Jon Shen reports on six months of breaking into their own agents.

Lorikeet’s Karla Burnett explains why making an agent vulnerable to exactly one attack is harder than making it secure. Jeffrey Aven covers the guardrail patterns for agents that touch production infrastructure. Dam Secure’s Simon Harloff explains how they built a security agent to review every pull request, from model choice to harnesses and evals. Adesh Gairola maps thousands of documented AI security incidents to the controls that stop them.

On the workshop day, Jeffrey shows how to build agents that safely operate real cloud infrastructure, and Simon how to build and evaluate a security agent.

Evidence over vibes

AI engineering is becoming an empirical discipline. Again and again, teams described replacing gut feel with evidence. Zendesk gives every agent a flight recorder, so a team can tell whether a new prompt, model or tool actually made things better.

Lorikeet’s Donna Zhou shows how a systematic eval flywheel raised the quality of a complex agent. Yulia Kuchina covers how to change an LLM system without guessing. Canva’s Dean Soste shares what 200,000 tool calls taught them about agentic evals.

Context is the new craft

What you put in front of the model, and what you leave out, has become a discipline of its own. Our first keynote speaker, Drew Breunig, is one of the people shaping how we think about context engineering, and he also runs a hands-on workshop on it.

Wesfarmers’ Gareth Williams makes the case for ontologies as AI’s operating manual for your business. Atlassian and Heidi show what memory looks like when the stakes are real.

The bottleneck has moved from writing code to everything around it

Coding agents can write code faster than teams can land it. Atlassian’s Shrey Somaiya shows what that means in a frontend codebase of more than 15 million lines. Culture Amp’s Artem Yakimenko shows that complexity now shows up directly on your API bill.

Deloitte’s Lee Simpson and Ethan McDonnell spent thirteen months on a government application with no hand-written code, and found their test strategy assumed humans wrote it. Cardiobase’s Mark Johnson borrows the systems engineering V-model to make software factories trustworthy.

High-stakes and regulated domains are moving first

Some of the most interesting work is happening where mistakes matter most. Skip’s Kevin Nguyen describes bringing LLMs into home loan credit decisions, with sign-off from the risk team.

V2 AI’s Stephen Sennett found that compliance asks less of engineers than many organisations assume. Ez Herrmann shares what it takes to build AI for students who can’t tell you when it’s wrong.

Leading and managing: the human factor

Not all of it is technical. Some of the most important questions are about people: how teams work, how they’re led, and how careers are changing. Andrew Murphy argues that teams aren’t resisting AI, they’re grieving their craft, and that the grief follows a predictable pattern. Inga Pflaumer says management is dead, but leadership is not.

Canva’s Sandra Arato shows how design and engineering ship together in an AI-native team. Lookahead’s Fiona Chan draws on Lookahead’s 2026 research into how we work, hire and grow now. You’ll find these talks and more under teams, leadership and work.

Coming up

Over the coming weeks we’ll go deeper:

• the key themes the programme is built around, including building agents that work in production, earning the right to trust them, and how software gets made
• the workshop day
• the tracks, once they’re finalised

In the meantime, you can explore the whole programme. This only scratches the surface of what will be around a hundred talks.

Join us in December

AI Engineer Sydney is 7 and 8 December at the Hilton Sydney, with a workshop day on 9 December. Every AI Engineer so far has sold out, and Melbourne went weeks in advance.

Register for AI Engineer Sydney. Bringing a team? Team passes include a complimentary Leadership upgrade for every five people.