Monday December 7
Registration
Doors open at 8am. Collect your badge, grab a coffee and meet our exhibitors before the opening keynotes.
Welcome to AI Engineer Sydney
John Allsopp opens AI Engineer Sydney and sets the scene for two days of engineering, leadership and research.
Prompts Rot. Keep the Source.
AI agents learn what we want from our feedback, then compress that intent into prompts, skills and memories—and discard the evidence. The result is a collection of natural-language artefacts that decay as models change…
The Harness Is the Agent
The model gets the headlines, but the harness is what turns a model into an agent. In late 2026 the harness is where the real engineering, and the real differentiation, lives. This keynote covers why the harness matters…
The BAML Programming Language
Whether you like it or not, it’s no longer possible to read all the code that’s generated by a model. The solution cannot be, "Just ship it," or you end up with slop everywhere. Neither can the solution be "Read…
The Elephant and the Goldfish: Architecture Patterns for Cutting 70% of Agent Token Costs in Production
LLM providers sell you a 2-million-token context window like an elephant that never forgets. If you actually build production agents that way, your latency explodes, your retrieval drifts, and your CFO will shut you…
Hackathon finalists and audience vote
The hackathon finalists present their work, followed by details of the audience vote.
Lunch
Lessons from Economics for the Human / Agent Software Workforce
AI is reshaping software engineering, creating uncertainty about how engineering roles will change as increasingly capable systems take on work that was once performed by people. While no one can predict the future with…
Ontologies: AI’s Operating Manual For Your Business
Tools churn. Factories commoditise. When everyone has the same models, the advantage goes to whoever gives their agents the clearest blueprint of how the business works - what exists, how it connects and the rules it…
Shipping together in an AI native team
While building an AI agent for marketing workflows at Leonardo.Ai, we discovered that the traditional design-to-engineering handoff was failing. A static mock could describe a happy path, but not how an agent would…
Lessons from Amazon's AI Journey
Learn how AWS transformed for and with AI with lessons learn around strategy, operating model, data foundations, governance, measurement, and change management.
Dependency hell is back. This time it's your agent's config.
Everyone has an agentic harness now. Ours is not special. What nobody has solved is running one across a whole team without every engineer drifting into a private setup.
We run Claude Code across 50 engineers and…
The CLI is dead, long live the CLI
Coding agents didn't kill the CLI. They became its most demanding users: they can't see a spinner, stall on prompts they can't answer, and treat error messages as instructions.
So we rewrote Sentry's CLI from scratch…
Your website is a terrible API: serving agents a different page at the edge
Every enterprise site we work on was built for a human with a browser. When an AI agent fetches the same URL it receives navigation, cookie banners, client-rendered components and marketing prose, then has to guess at…
Sovereignty Is an Inference Problem
AI sovereignty isn’t about where the GPUs sit. It’s about who can switch the intelligence off.
Australia is spending billions on sovereign AI infrastructure. But onshore compute running models subject to another…
My Agents broke APIs - Fixing Multi-Agent Systems with MCP
Modern AI agents struggle not because of reasoning limits, but because of interaction with tools on interfaces designed for humans. In agentic systems, this mismatch leads to incorrect tool selection, redundant calls,…
One harness for every coding agent: swap models and runtimes without rewriting
You want your team using the best coding agent for each job : Claude Code, Codex, Cursor, or one you built yourself , without rewiring your setup every time a better model ships, and without each agent running loose…
Inside Dovetail's Secure Software Factory: risk-tiered code review when everyone ships
At Dovetail, engineers, designers, product managers and even the CEO ship code. Weekly merged pull requests have tripled since January, and risk-tiered review has cut time spent waiting for review by 70%.
In this joint…
Why software factories can't be trusted (and how the Systems Engineering V-model helps)
Coding agents will do anything to get to a PR. Ours marked tasks complete that didn't exist, wrote "next steps: verification" under unreviewed pull requests, and quietly skipped every boring stage between the ticket and…
Token usage is the new lines of code (and it's just as useless)
Canva hit mid-2026 with every AI metric a vendor could hand us and no answer to the only question the business actually cared about: is this working? Adoption flattened at around 85% of ~3,000 engineers, and the number…
We Deleted Most of Our Agents. Everything Got Faster
The default advice is to add agents: specialised roles, an orchestrator, handoffs between them. We built that. It was a genuinely useful way to explore the problem space, and under real production traffic it was slow…
Applied Decision Models: A Closer Look at System One Models, and How and Where to Use Them
In this session, I'll look at so-called "System One" models: how they work at a high level, and the use cases they're best suited to.
I'll use two Cloudflare models as examples. The first is clef, an open-weight model…
The Guess Never Becomes a Fact: Provenance-Governed, Model-Free Memory for LLMs
Long context solves within-session coherence, not cross-session persistence. Once an assistant starts storing memories, a quieter failure appears: its own inferences can be written back, retrieved in later sessions, and…
Don't stop, won't stop. Automated verification of long-running agentic loops
Long-running agentic loops create different engineering problems than code generation. Once agents work unattended for many hours, iterate through 20 or more review cycles, and build stacked PRs towards a larger goal,…
Running AI Coding Agents at Scale: Adopting New Models, Controlling Cost, and Governing What Agents Do
Running coding agents across a company comes down to three questions: which models to offer, how to control cost, and what agents should be allowed to do. This talk gives four recommendations on those questions, drawn…
In conversation
The session’s three speakers join the MC for a conversation.
It's time for a new kind of software
Applications package a predetermined interface, data model and set of capabilities around somebody else’s idea of a task. Generative models can now produce software on demand, but generative UI today either sits atop…
Less is More: Optimising Search for Agents
Giving agents more search results doesn't necessarily make them better.
Ryan Ahern explains how Exa is rethinking search for agents, improving task success rates while reducing the cost of finding the information they…
The Machine That Changed the World, vAI
AI agents can write code faster than any team can review it. Code is abundant. Confidence is scarce. Join Andre to learn how leading enterprises are building agentic software factories they can trust. Learn about:
-…
Afternoon break
Your spaghetti code has an invoice now: what complexity does to coding agents
We've told engineers for decades that high complexity makes code harder for humans to reason about. It turns out it makes code measurably more expensive for agents too. Unlike human frustration, this shows up directly…
The Model Wasn't the MOAT: How 10 Design System Engineers Turned Platform Knowledge into Enterprise AI
Most enterprise AI strategies begin with models, central AI teams, and a list of potential use cases. We started somewhere else: with a platform organisation that already shaped how more than 2,000 frontend engineers…
Tools Before Autonomy: What 200,000 Tool Calls Taught Us About Agentic Evals
Production AI problems are rarely observable through deterministic CI checks. They surface as user feedback, bad LLM-as-a-Judge scores, alerts, or even a vague sense that something is wrong. The evidence that explains…
MCP Isn't Enough: Why Agents Needed an Event Loop, Not Just an API
Many apps are adding MCP interfaces so agents can query their systems of record. It works, until the workflows we actually care about turn out to be "something happened, now respond", and polling wastes network, context…
Your Coding Agent Is Fast. Your Codebase Is the Bottleneck.
AI agents can write code faster than teams can land it. Jira’s frontend codebase contains over 15 million lines of code, supports thousands of contributors, and doubles in size roughly every two years.
At this scale,…
How to Change an LLM System Without Guessing
We were running a production LLM pipeline that classified legal documents, and every change was a guess. Swap a prompt, change a model — better or worse? Nobody could say. The outputs looked plausible either way, and…
An AI-Assisted Brand Uplift at Scale
What if designers could build directly in the codebase? We used AI-assisted workflows to let designers prototype in production code with live preview links, while engineers focused on feasibility and scale. The result:…
From REST to Agentic: Trusted APIs in the age of AI
Enterprise platforms don't get to start from scratch. When AI agents need to act on behalf of users, they inherit the API surfaces those users already depend on — surfaces evolved over many years for human-driven…
AI Janitor: Making Architecture Review Executable for Coding Agents
AI coding agents accelerate code generation, but they also accelerate architectural drift, regressions and false confidence. While building a multi-service production platform with only two of us, I found that a normal…
From vibes to a systematic eval flywheel: evals for high stakes AI agents
If you are sceptical about how evals drive real results, or looking for more depth than introductory tutorial videos, this talk is for you.
We'll show you how Lorikeet, an AI customer service startup, built an eval…
Ship AI Agents Faster, from Prototype to Production
As AI agents take on more autonomous roles, the identity and access patterns built for humans and APIs no longer hold. This session draws on Auth0’s real-world experience securing agentic workflows where misconfigured…
209 ports in six days, in a language the model barely knew
Jolt and jank are two young Clojure implementations, one running on Chez Scheme and one compiling to native code through C++/LLVM. Between them they have almost no public code for a model to have learned from. Ask…
Where Should the Dice Roll? Placing Non-Determinism Deliberately in Enterprise AI
Most production AI guidance assumes you want consistency: pin the prompt, lower the temperature, eval for drift. But a whole class of enterprise use cases - idea generation, recommendation, exploration, synthesis - is…
Does This Agent Make My Context Look Big? Right-Sizing AI Architectures for Production
One agent that does everything is powerful, until it’s impossible to debug, eval or secure. Break it into a graph of specialised agents and you get control, until it doesn't for a multitude of reasons.
Neither…
Talk to be confirmed
This programme position is being held for an Industry presentation to be announced.
Compounding lessons into skills: How we performed a large code migration using AI (with no new bugs!)
Canva's "Ingredient Generation" service powers all of our media generation experiences. Every image, video, audio and 3d object request for generation goes through this service, which offers 50+ AI Models. One of its…
Define Good, Then Ship: Blind Evals for Agentic Features
In May I rebuilt our brand-strategy pipeline as a fully agentic system, replacing a chain of LLM calls with a managed agent. I compared old and new across 44 customers, and the new system won every time. But I was the…
Talk to be confirmed
This programme position is being held for an Industry presentation to be announced.
In conversation
The session’s three speakers join the MC for a conversation.
Your Test Strategy Assumes Humans Write Code
Thirteen months building a custom state government full-stack application with zero code written by hand.
7,137 tests, 5,631 of them unit tests. It was a textbook testing pyramid but still bugs stacked up.
The…
The agents went rogue at 2%
Giving an agent a skill sounds straightforward: it loads a Markdown file, follows the instructions and completes the task.
Ours launched a long-running, paid process and told the agent to wait for the result. But…
Talk to be confirmed
This programme position is being held for an Industry presentation to be announced.
Reception
At the close of day one, join fellow attendees and speakers for drinks, conversation and the chance to continue the day’s discussions.
Speaker dinner
Leadership ticket holders join the speakers for dinner after the reception. Final venue details will be shared with guests.
Registration
Doors open at 8am. Collect your badge, grab a coffee and meet our exhibitors before the opening keynotes.
Welcome to AI Engineer Sydney
John Allsopp opens AI Engineer Sydney and sets the scene for two days of engineering, leadership and research.
Prompts Rot. Keep the Source.
AI agents learn what we want from our feedback, then compress that intent into prompts, skills and memories—and discard the evidence. The result is a collection of natural-language artefacts that decay as models change…
The Harness Is the Agent
The model gets the headlines, but the harness is what turns a model into an agent. In late 2026 the harness is where the real engineering, and the real differentiation, lives. This keynote covers why the harness matters…
The BAML Programming Language
Whether you like it or not, it’s no longer possible to read all the code that’s generated by a model. The solution cannot be, "Just ship it," or you end up with slop everywhere. Neither can the solution be "Read…
The Elephant and the Goldfish: Architecture Patterns for Cutting 70% of Agent Token Costs in Production
LLM providers sell you a 2-million-token context window like an elephant that never forgets. If you actually build production agents that way, your latency explodes, your retrieval drifts, and your CFO will shut you…
Hackathon finalists and audience vote
The hackathon finalists present their work, followed by details of the audience vote.
Lunch
Lessons from Economics for the Human / Agent Software Workforce
AI is reshaping software engineering, creating uncertainty about how engineering roles will change as increasingly capable systems take on work that was once performed by people. While no one can predict the future with…
Ontologies: AI’s Operating Manual For Your Business
Tools churn. Factories commoditise. When everyone has the same models, the advantage goes to whoever gives their agents the clearest blueprint of how the business works - what exists, how it connects and the rules it…
Shipping together in an AI native team
While building an AI agent for marketing workflows at Leonardo.Ai, we discovered that the traditional design-to-engineering handoff was failing. A static mock could describe a happy path, but not how an agent would…
Lessons from Amazon's AI Journey
Learn how AWS transformed for and with AI with lessons learn around strategy, operating model, data foundations, governance, measurement, and change management.
Your website is a terrible API: serving agents a different page at the edge
Every enterprise site we work on was built for a human with a browser. When an AI agent fetches the same URL it receives navigation, cookie banners, client-rendered components and marketing prose, then has to guess at…
Dependency hell is back. This time it's your agent's config.
Everyone has an agentic harness now. Ours is not special. What nobody has solved is running one across a whole team without every engineer drifting into a private setup.
We run Claude Code across 50 engineers and…
The CLI is dead, long live the CLI
Coding agents didn't kill the CLI. They became its most demanding users: they can't see a spinner, stall on prompts they can't answer, and treat error messages as instructions.
So we rewrote Sentry's CLI from scratch…
Sovereignty Is an Inference Problem
AI sovereignty isn’t about where the GPUs sit. It’s about who can switch the intelligence off.
Australia is spending billions on sovereign AI infrastructure. But onshore compute running models subject to another…
My Agents broke APIs - Fixing Multi-Agent Systems with MCP
Modern AI agents struggle not because of reasoning limits, but because of interaction with tools on interfaces designed for humans. In agentic systems, this mismatch leads to incorrect tool selection, redundant calls,…
One harness for every coding agent: swap models and runtimes without rewriting
You want your team using the best coding agent for each job : Claude Code, Codex, Cursor, or one you built yourself , without rewiring your setup every time a better model ships, and without each agent running loose…
Inside Dovetail's Secure Software Factory: risk-tiered code review when everyone ships
At Dovetail, engineers, designers, product managers and even the CEO ship code. Weekly merged pull requests have tripled since January, and risk-tiered review has cut time spent waiting for review by 70%.
In this joint…
Token usage is the new lines of code (and it's just as useless)
Canva hit mid-2026 with every AI metric a vendor could hand us and no answer to the only question the business actually cared about: is this working? Adoption flattened at around 85% of ~3,000 engineers, and the number…
We Deleted Most of Our Agents. Everything Got Faster
The default advice is to add agents: specialised roles, an orchestrator, handoffs between them. We built that. It was a genuinely useful way to explore the problem space, and under real production traffic it was slow…
Why software factories can't be trusted (and how the Systems Engineering V-model helps)
Coding agents will do anything to get to a PR. Ours marked tasks complete that didn't exist, wrote "next steps: verification" under unreviewed pull requests, and quietly skipped every boring stage between the ticket and…
Applied Decision Models: A Closer Look at System One Models, and How and Where to Use Them
In this session, I'll look at so-called "System One" models: how they work at a high level, and the use cases they're best suited to.
I'll use two Cloudflare models as examples. The first is clef, an open-weight model…
The Guess Never Becomes a Fact: Provenance-Governed, Model-Free Memory for LLMs
Long context solves within-session coherence, not cross-session persistence. Once an assistant starts storing memories, a quieter failure appears: its own inferences can be written back, retrieved in later sessions, and…
Don't stop, won't stop. Automated verification of long-running agentic loops
Long-running agentic loops create different engineering problems than code generation. Once agents work unattended for many hours, iterate through 20 or more review cycles, and build stacked PRs towards a larger goal,…
Running AI Coding Agents at Scale: Adopting New Models, Controlling Cost, and Governing What Agents Do
Running coding agents across a company comes down to three questions: which models to offer, how to control cost, and what agents should be allowed to do. This talk gives four recommendations on those questions, drawn…
In conversation
The session’s three speakers join the MC for a conversation.
Less is More: Optimising Search for Agents
Giving agents more search results doesn't necessarily make them better.
Ryan Ahern explains how Exa is rethinking search for agents, improving task success rates while reducing the cost of finding the information they…
It's time for a new kind of software
Applications package a predetermined interface, data model and set of capabilities around somebody else’s idea of a task. Generative models can now produce software on demand, but generative UI today either sits atop…
The Machine That Changed the World, vAI
AI agents can write code faster than any team can review it. Code is abundant. Confidence is scarce. Join Andre to learn how leading enterprises are building agentic software factories they can trust. Learn about:
-…
Afternoon break
The Model Wasn't the MOAT: How 10 Design System Engineers Turned Platform Knowledge into Enterprise AI
Most enterprise AI strategies begin with models, central AI teams, and a list of potential use cases. We started somewhere else: with a platform organisation that already shaped how more than 2,000 frontend engineers…
Tools Before Autonomy: What 200,000 Tool Calls Taught Us About Agentic Evals
Production AI problems are rarely observable through deterministic CI checks. They surface as user feedback, bad LLM-as-a-Judge scores, alerts, or even a vague sense that something is wrong. The evidence that explains…
Your spaghetti code has an invoice now: what complexity does to coding agents
We've told engineers for decades that high complexity makes code harder for humans to reason about. It turns out it makes code measurably more expensive for agents too. Unlike human frustration, this shows up directly…
MCP Isn't Enough: Why Agents Needed an Event Loop, Not Just an API
Many apps are adding MCP interfaces so agents can query their systems of record. It works, until the workflows we actually care about turn out to be "something happened, now respond", and polling wastes network, context…
How to Change an LLM System Without Guessing
We were running a production LLM pipeline that classified legal documents, and every change was a guess. Swap a prompt, change a model — better or worse? Nobody could say. The outputs looked plausible either way, and…
Your Coding Agent Is Fast. Your Codebase Is the Bottleneck.
AI agents can write code faster than teams can land it. Jira’s frontend codebase contains over 15 million lines of code, supports thousands of contributors, and doubles in size roughly every two years.
At this scale,…
An AI-Assisted Brand Uplift at Scale
What if designers could build directly in the codebase? We used AI-assisted workflows to let designers prototype in production code with live preview links, while engineers focused on feasibility and scale. The result:…
From REST to Agentic: Trusted APIs in the age of AI
Enterprise platforms don't get to start from scratch. When AI agents need to act on behalf of users, they inherit the API surfaces those users already depend on — surfaces evolved over many years for human-driven…
From vibes to a systematic eval flywheel: evals for high stakes AI agents
If you are sceptical about how evals drive real results, or looking for more depth than introductory tutorial videos, this talk is for you.
We'll show you how Lorikeet, an AI customer service startup, built an eval…
AI Janitor: Making Architecture Review Executable for Coding Agents
AI coding agents accelerate code generation, but they also accelerate architectural drift, regressions and false confidence. While building a multi-service production platform with only two of us, I found that a normal…
Ship AI Agents Faster, from Prototype to Production
As AI agents take on more autonomous roles, the identity and access patterns built for humans and APIs no longer hold. This session draws on Auth0’s real-world experience securing agentic workflows where misconfigured…
Does This Agent Make My Context Look Big? Right-Sizing AI Architectures for Production
One agent that does everything is powerful, until it’s impossible to debug, eval or secure. Break it into a graph of specialised agents and you get control, until it doesn't for a multitude of reasons.
Neither…
Where Should the Dice Roll? Placing Non-Determinism Deliberately in Enterprise AI
Most production AI guidance assumes you want consistency: pin the prompt, lower the temperature, eval for drift. But a whole class of enterprise use cases - idea generation, recommendation, exploration, synthesis - is…
209 ports in six days, in a language the model barely knew
Jolt and jank are two young Clojure implementations, one running on Chez Scheme and one compiling to native code through C++/LLVM. Between them they have almost no public code for a model to have learned from. Ask…
Talk to be confirmed
This programme position is being held for an Industry presentation to be announced.
Define Good, Then Ship: Blind Evals for Agentic Features
In May I rebuilt our brand-strategy pipeline as a fully agentic system, replacing a chain of LLM calls with a managed agent. I compared old and new across 44 customers, and the new system won every time. But I was the…
Compounding lessons into skills: How we performed a large code migration using AI (with no new bugs!)
Canva's "Ingredient Generation" service powers all of our media generation experiences. Every image, video, audio and 3d object request for generation goes through this service, which offers 50+ AI Models. One of its…
Talk to be confirmed
This programme position is being held for an Industry presentation to be announced.
In conversation
The session’s three speakers join the MC for a conversation.
The agents went rogue at 2%
Giving an agent a skill sounds straightforward: it loads a Markdown file, follows the instructions and completes the task.
Ours launched a long-running, paid process and told the agent to wait for the result. But…
Your Test Strategy Assumes Humans Write Code
Thirteen months building a custom state government full-stack application with zero code written by hand.
7,137 tests, 5,631 of them unit tests. It was a textbook testing pyramid but still bugs stacked up.
The…
Talk to be confirmed
This programme position is being held for an Industry presentation to be announced.
Reception
At the close of day one, join fellow attendees and speakers for drinks, conversation and the chance to continue the day’s discussions.
Speaker dinner
Leadership ticket holders join the speakers for dinner after the reception. Final venue details will be shared with guests.
Tuesday December 8
Leadership Breakfast with Notion: The Lonely Builder — Thinking Together in the Age of AI
A breakfast for engineering and technology leaders, hosted by Notion. Places are limited: join the waitlist to attend.
As coding agents become everyday collaborators, engineers can move faster, but may find themselves…
Welcome to day two
John Allsopp opens the second day of AI Engineer Sydney.
Keynote speaker to be announced
This keynote position is being held while one further speaker is finalised.
The State of Open Source AI
Raffi Krikorian, Chief Technology Officer at Mozilla, explores the state of open-source AI, with a preview of Mozilla’s next State of Open Source AI report.
Who Keeps It Open?
Every generational shift in computing produces plumbing that everyone depends on. TCP/IP, HTTP, the container runtime. And every time, there is a window where someone can quietly come to own it, and everyone else pays…
Keynote speaker to be announced
One further editorial keynote will be announced for this position.
The AI year: how we work, hire and grow now
For a lot of teams, this was the year AI stopped being an experiment and started being part of the job. Shipping got faster, but so did the pace, the pressure and the pile of code waiting for review.
Drawing on…
Hackathon winner and closing
We announce the hackathon winner and close the keynote programme before the day’s tracks begin.
Lunch
Oops I Hacked It Again: Six Months of Red Teaming Our AI Agents
In March I onboarded our first team member dedicated to AI Red Teaming. That was a month before the announcement of Claude Mythos Preview, and the open letters that followed from APRA in April and ASIC in May…
How we introduced LLMs into credit decisioning
Over the past nine months we've built and deployed an AI-assisted credit assessment system that's now used in production to help credit assessors process home loan applications. More importantly, it was signed off by…
The Benchmark Ends. The World Doesn’t: Building Persistent Engineering Environments for Continual Learning
Large language models can complete sequences of tasks—but continual learning is not just doing more tasks. Most agent environments reset after every episode: state disappears, required follow-up vanishes, delayed…
The Contradictions of Building AI Products and Teams
For every claim about how to build with AI, there seems to be an equal and opposite claim that’s also true. Chat is the universal interface, so go headless. Never bolt AI onto your product, reimagine it. Everyone on…
Deliberate Holes Only
Making an AI agent vulnerable to exactly one attack is harder than making it completely secure, let alone doing it six different times.
In this talk, I'll describe the process of building an agentic AI capture the…
Extracting Tacit Knowledge for Production Agents
An eval is a definition of "good" written down well enough for a machine to grade against. In a real business, "good" isn't written down anywhere. It lives in someone's reaction, in what past work actually did, and in…
Sovereign AI Across the Stack: From Infrastructure to Inference
Building sovereign AI takes more than Australian-hosted infrastructure. It requires control across the entire AI stack, from GPU compute and model deployment to inference, APIs and the platform connecting it all.…
Building AI for students who can't tell you when it's wrong
Teachers in Australian specialist schools spend two hours or more preparing a single lesson, then adapt it again for every ability group in the room. Many are already pasting student details into consumer chatbots to…
Query before mutation: guardrail patterns for agents that touch production infrastructure
Giving an AI agent permission to change production infrastructure removes the human pause that made plan-and-apply safe. In this talk, Jeffrey Aven shows how to put that judgement back into the system itself.
He…
Adaptive Learning: Encoding Clinician Edits as Memory
Clinicians edit their AI-generated notes for many reasons: adding or removing content, fixing spelling, and preferences for structure, wording, or terminology. Some of these edits are contextual, applying in some…
We Replaced Our AI Framework in Two Weeks
Our main product is a conversational agent that helps users explore and analyse their data. As LLM providers released capabilities such as finer-grained cache checkpoints and tool search, the framework underneath our…
Building Agentic Memory for an AI SOC: Why Our Semantic Cache Was the Wrong Answer
We run a fleet of AI agents against production security detections at Atlassian. One tunes noisy detection rules, another reviews new ones, and more are coming for alert triage and hunting. Every one of them started…
Regulated Doesn't Mean On Rails: what compliance actually asks of your engineers
We spent three months rebuilding a regulated enterprise's development lifecycle around agents, then worked out how much of the governance we had added was doing real compliance work. Less than we assumed.
Across…
AI Sandboxes: Running Coding Agents Safely in Production-Grade Environments
The number of cyber attacks and security risks related to Coding Agents has sky rocketed. AI coding agents like Claude Code, Codex CLI, and Gemini CLI don’t behave like your typical developer tools. They install system…
Dark Software Factory: Why We Build Differently
What happens when traditional software development approaches can no longer solve the problems we face? In this session, we explore the challenges, gaps, and limitations that remain unsolved, why conventional approaches…
From Documents to Defensible Evidence: Rethinking RAG for Audits
Audits impose strict requirements for evidence, traceability and expert judgement. Each entity being audited supplies a new body of evidence, made up of manuals, records, tables, diagrams, scans and multilingual…
Beyond System Prompts: Securing AI Agents with Fine-Grained Authorization
To make an AI agent useful, we often give it broad access to APIs and data. As systems move from simple retrieval to autonomous action, system-prompt instructions are not an adequate security boundary.
This…
Talk to be confirmed
This programme position is being held for a presentation to be announced.
In conversation
The session’s three speakers join the MC for a conversation.
Building a Security Agent: Model choice, Harnesses and Evals
In this presentation, I’ll show how we set out to give developers useful security feedback on every pull request in under three minutes. The benchmark results and methodology are published here:…
Classifiers are dead. Long live classifiers!
Training a classifier used to mean collecting labelled data, choosing an architecture, training a model, evaluating it, and deploying it. Today, for a surprising number of problems, you can replace most of that with a…
Talk to be confirmed
This programme position is being held for a presentation to be announced.
Afternoon break
Running 100K+ Concurrent Agents: It's Never the Prompt
Your Agents Aren't Failing at Prompting
Every agent demo works. Then you run ten thousand at once, and the failures have nothing to do with the model: dead tuples piling up, a sync HTTP client freezing the event…
Trust is engineered, not granted: why we focus on verifying before background coding agents
Heidi is an AI scribe used by 130K clinicians a week. Our 150 engineers ship 100+ PRs daily into prod, and AI made writing code so cheap that review became the bottleneck: our P75 review wait was 14 hours, almost all of…
Management Is Dead, Leadership Is Not
Engineering teams are getting smaller. A team of four with strong AI tooling now ships what took twelve people two years ago, and the org chart has not caught up. The first layer to feel it is middle management - the…
Skippy: The Little Inference Engine That CUDA
What happens under the hood when you run a large language model? This talk explores Skippy, a custom inference engine built for Mesh on top of llama.cpp, and the engineering behind local and distributed AI inference.…
From Prompt Rules to Structural Guarantees: The Harness Behind a Production Analytics Agent
Checkout AI answers open-ended questions about retail sales data in natural language. It plans,
calls analytics tools over MCP, executes Python in a sandbox, and returns a written analysis with
charts. In a system…
Don't Fight Hallucinations. Make Them Impossible
The Heatseeker AI chat answers data questions for marketers who make decisions with million-dollar budgets. Wrong answers or hallucinated numbers are not an option here, as you can imagine ;)
The fight against them…
Local AI and the End of the Cloud Default
Most AI today runs in the cloud, where every token costs money and every input leaves your device. Now AI agents are running around the clock with access to your email, files, and most sensitive data. That burns through…
Half our handoffs are the bot's idea
About one in four conversations with WordPress.com's AI support bot ends with a person. In about half of those, the bot suggested it.
These are two different problems. In one, the user asks for a person. In the…
Give Every Agent a Flight Recorder
Agent teams should not need to file a ticket with a central evaluation team just to learn whether a new prompt, model, or tool made their agent better. At Zendesk, we developed and deployed a trace-first evaluation…
Safely releasing AI applications while continuously patching vulnerabilities
AI applications still need regular security patches and dependency updates, but those software changes can also affect the data processing, model interaction, and evaluation paths around an LLM. In our releases,…
Robot Kintsugi: The Art and Design of Reengineering Old Hardware
What does it take to give an old robot new life with AI?
This talk explores the art and design of reengineering abandoned hardware, treating its physical constraints and original character as starting points for new…
Office-Brain and the Night Shift: What Happens When You Turn Off the Meter
"Good enough" AI now runs on a reasonably beefy laptop. Qwen3.8-27B benchmarks at an Intelligence Index of 52 - roughly the best money could buy back in February - and inferences happily on consumer kit. When tokens are…
Should You Build a Software Factory? Tales from an Open-Source Maintainer
Coding agents helped me write code faster, but I kept running into problems elsewhere in my engineering workflow. Tests passed without proving much. Parallel tasks changed the same files. Reviews piled up. This talk…
The five stages of losing our craft
Your best engineer won't use AI tools. Your tech lead is using them but won't tell anyone. Your new grads don't understand what the fuss is about. And you, the leader, are supposed to have answers for all of them.…
How to Count to One Hundred
Agents don't work well together in chat rooms: put a few in a channel and they talk past each other. Natural language and unrestricted JSON feed an open-ended pipeline of reasoning and tool calls with full Turing power.…
Autonomous agents need autonomous protection
Autonomous agents need autonomous protection. An agent that acts alone at 3am cannot wait for a quarterly pen test, a consultant's report, or a security engineer the team does not have. We started with the data. We…
Talk to be confirmed
This programme position is being held for a presentation to be announced.
Talk to be confirmed
This programme position is being held for a Research / Deep Tech presentation to be announced.
In conversation
The session’s three speakers join the MC for a conversation.
Assert Chaos
Once a product reaches the wilds of production, all sorts of things can go wrong. Carefully logging and monitoring are one thing, but what if we could do better?
Bring new levels of code stability by recreating the…
Talk to be confirmed
This programme position is being held for a presentation to be announced.
Talk to be confirmed
This programme position is being held for a Research / Deep Tech presentation to be announced.
AI Engineer Sydney closing reception
Join fellow attendees and speakers from 6:00 pm for drinks, conversation and a final chance to continue the conference’s discussions together. Gadigal Station is just around the corner from Hilton Sydney; Blackwattle is…
Leadership Breakfast with Notion: The Lonely Builder — Thinking Together in the Age of AI
A breakfast for engineering and technology leaders, hosted by Notion. Places are limited: join the waitlist to attend.
As coding agents become everyday collaborators, engineers can move faster, but may find themselves…
Welcome to day two
John Allsopp opens the second day of AI Engineer Sydney.
Keynote speaker to be announced
This keynote position is being held while one further speaker is finalised.
The State of Open Source AI
Raffi Krikorian, Chief Technology Officer at Mozilla, explores the state of open-source AI, with a preview of Mozilla’s next State of Open Source AI report.
Who Keeps It Open?
Every generational shift in computing produces plumbing that everyone depends on. TCP/IP, HTTP, the container runtime. And every time, there is a window where someone can quietly come to own it, and everyone else pays…
Keynote speaker to be announced
One further editorial keynote will be announced for this position.
The AI year: how we work, hire and grow now
For a lot of teams, this was the year AI stopped being an experiment and started being part of the job. Shipping got faster, but so did the pace, the pressure and the pile of code waiting for review.
Drawing on…
Hackathon winner and closing
We announce the hackathon winner and close the keynote programme before the day’s tracks begin.
Lunch
How we introduced LLMs into credit decisioning
Over the past nine months we've built and deployed an AI-assisted credit assessment system that's now used in production to help credit assessors process home loan applications. More importantly, it was signed off by…
Oops I Hacked It Again: Six Months of Red Teaming Our AI Agents
In March I onboarded our first team member dedicated to AI Red Teaming. That was a month before the announcement of Claude Mythos Preview, and the open letters that followed from APRA in April and ASIC in May…
The Contradictions of Building AI Products and Teams
For every claim about how to build with AI, there seems to be an equal and opposite claim that’s also true. Chat is the universal interface, so go headless. Never bolt AI onto your product, reimagine it. Everyone on…
The Benchmark Ends. The World Doesn’t: Building Persistent Engineering Environments for Continual Learning
Large language models can complete sequences of tasks—but continual learning is not just doing more tasks. Most agent environments reset after every episode: state disappears, required follow-up vanishes, delayed…
Deliberate Holes Only
Making an AI agent vulnerable to exactly one attack is harder than making it completely secure, let alone doing it six different times.
In this talk, I'll describe the process of building an agentic AI capture the…
Sovereign AI Across the Stack: From Infrastructure to Inference
Building sovereign AI takes more than Australian-hosted infrastructure. It requires control across the entire AI stack, from GPU compute and model deployment to inference, APIs and the platform connecting it all.…
Extracting Tacit Knowledge for Production Agents
An eval is a definition of "good" written down well enough for a machine to grade against. In a real business, "good" isn't written down anywhere. It lives in someone's reaction, in what past work actually did, and in…
Building AI for students who can't tell you when it's wrong
Teachers in Australian specialist schools spend two hours or more preparing a single lesson, then adapt it again for every ability group in the room. Many are already pasting student details into consumer chatbots to…
Query before mutation: guardrail patterns for agents that touch production infrastructure
Giving an AI agent permission to change production infrastructure removes the human pause that made plan-and-apply safe. In this talk, Jeffrey Aven shows how to put that judgement back into the system itself.
He…
We Replaced Our AI Framework in Two Weeks
Our main product is a conversational agent that helps users explore and analyse their data. As LLM providers released capabilities such as finer-grained cache checkpoints and tool search, the framework underneath our…
Adaptive Learning: Encoding Clinician Edits as Memory
Clinicians edit their AI-generated notes for many reasons: adding or removing content, fixing spelling, and preferences for structure, wording, or terminology. Some of these edits are contextual, applying in some…
Regulated Doesn't Mean On Rails: what compliance actually asks of your engineers
We spent three months rebuilding a regulated enterprise's development lifecycle around agents, then worked out how much of the governance we had added was doing real compliance work. Less than we assumed.
Across…
AI Sandboxes: Running Coding Agents Safely in Production-Grade Environments
The number of cyber attacks and security risks related to Coding Agents has sky rocketed. AI coding agents like Claude Code, Codex CLI, and Gemini CLI don’t behave like your typical developer tools. They install system…
Dark Software Factory: Why We Build Differently
What happens when traditional software development approaches can no longer solve the problems we face? In this session, we explore the challenges, gaps, and limitations that remain unsolved, why conventional approaches…
Building Agentic Memory for an AI SOC: Why Our Semantic Cache Was the Wrong Answer
We run a fleet of AI agents against production security detections at Atlassian. One tunes noisy detection rules, another reviews new ones, and more are coming for alert triage and hunting. Every one of them started…
Beyond System Prompts: Securing AI Agents with Fine-Grained Authorization
To make an AI agent useful, we often give it broad access to APIs and data. As systems move from simple retrieval to autonomous action, system-prompt instructions are not an adequate security boundary.
This…
Talk to be confirmed
This programme position is being held for a presentation to be announced.
From Documents to Defensible Evidence: Rethinking RAG for Audits
Audits impose strict requirements for evidence, traceability and expert judgement. Each entity being audited supplies a new body of evidence, made up of manuals, records, tables, diagrams, scans and multilingual…
In conversation
The session’s three speakers join the MC for a conversation.
Building a Security Agent: Model choice, Harnesses and Evals
In this presentation, I’ll show how we set out to give developers useful security feedback on every pull request in under three minutes. The benchmark results and methodology are published here:…
Talk to be confirmed
This programme position is being held for a presentation to be announced.
Classifiers are dead. Long live classifiers!
Training a classifier used to mean collecting labelled data, choosing an architecture, training a model, evaluating it, and deploying it. Today, for a surprising number of problems, you can replace most of that with a…
Afternoon break
Management Is Dead, Leadership Is Not
Engineering teams are getting smaller. A team of four with strong AI tooling now ships what took twelve people two years ago, and the org chart has not caught up. The first layer to feel it is middle management - the…
Running 100K+ Concurrent Agents: It's Never the Prompt
Your Agents Aren't Failing at Prompting
Every agent demo works. Then you run ten thousand at once, and the failures have nothing to do with the model: dead tuples piling up, a sync HTTP client freezing the event…
Trust is engineered, not granted: why we focus on verifying before background coding agents
Heidi is an AI scribe used by 130K clinicians a week. Our 150 engineers ship 100+ PRs daily into prod, and AI made writing code so cheap that review became the bottleneck: our P75 review wait was 14 hours, almost all of…
Skippy: The Little Inference Engine That CUDA
What happens under the hood when you run a large language model? This talk explores Skippy, a custom inference engine built for Mesh on top of llama.cpp, and the engineering behind local and distributed AI inference.…
Don't Fight Hallucinations. Make Them Impossible
The Heatseeker AI chat answers data questions for marketers who make decisions with million-dollar budgets. Wrong answers or hallucinated numbers are not an option here, as you can imagine ;)
The fight against them…
From Prompt Rules to Structural Guarantees: The Harness Behind a Production Analytics Agent
Checkout AI answers open-ended questions about retail sales data in natural language. It plans,
calls analytics tools over MCP, executes Python in a sandbox, and returns a written analysis with
charts. In a system…
Local AI and the End of the Cloud Default
Most AI today runs in the cloud, where every token costs money and every input leaves your device. Now AI agents are running around the clock with access to your email, files, and most sensitive data. That burns through…
Half our handoffs are the bot's idea
About one in four conversations with WordPress.com's AI support bot ends with a person. In about half of those, the bot suggested it.
These are two different problems. In one, the user asks for a person. In the…
Give Every Agent a Flight Recorder
Agent teams should not need to file a ticket with a central evaluation team just to learn whether a new prompt, model, or tool made their agent better. At Zendesk, we developed and deployed a trace-first evaluation…
Safely releasing AI applications while continuously patching vulnerabilities
AI applications still need regular security patches and dependency updates, but those software changes can also affect the data processing, model interaction, and evaluation paths around an LLM. In our releases,…
Robot Kintsugi: The Art and Design of Reengineering Old Hardware
What does it take to give an old robot new life with AI?
This talk explores the art and design of reengineering abandoned hardware, treating its physical constraints and original character as starting points for new…
The five stages of losing our craft
Your best engineer won't use AI tools. Your tech lead is using them but won't tell anyone. Your new grads don't understand what the fuss is about. And you, the leader, are supposed to have answers for all of them.…
Office-Brain and the Night Shift: What Happens When You Turn Off the Meter
"Good enough" AI now runs on a reasonably beefy laptop. Qwen3.8-27B benchmarks at an Intelligence Index of 52 - roughly the best money could buy back in February - and inferences happily on consumer kit. When tokens are…
Should You Build a Software Factory? Tales from an Open-Source Maintainer
Coding agents helped me write code faster, but I kept running into problems elsewhere in my engineering workflow. Tests passed without proving much. Parallel tasks changed the same files. Reviews piled up. This talk…
How to Count to One Hundred
Agents don't work well together in chat rooms: put a few in a channel and they talk past each other. Natural language and unrestricted JSON feed an open-ended pipeline of reasoning and tool calls with full Turing power.…
Talk to be confirmed
This programme position is being held for a presentation to be announced.
Autonomous agents need autonomous protection
Autonomous agents need autonomous protection. An agent that acts alone at 3am cannot wait for a quarterly pen test, a consultant's report, or a security engineer the team does not have. We started with the data. We…
Talk to be confirmed
This programme position is being held for a Research / Deep Tech presentation to be announced.
In conversation
The session’s three speakers join the MC for a conversation.
Talk to be confirmed
This programme position is being held for a presentation to be announced.
Assert Chaos
Once a product reaches the wilds of production, all sorts of things can go wrong. Carefully logging and monitoring are one thing, but what if we could do better?
Bring new levels of code stability by recreating the…
Talk to be confirmed
This programme position is being held for a Research / Deep Tech presentation to be announced.
AI Engineer Sydney closing reception
Join fellow attendees and speakers from 6:00 pm for drinks, conversation and a final chance to continue the conference’s discussions together. Gadigal Station is just around the corner from Hilton Sydney; Blackwattle is…
Wednesday December 9
Build and Evaluate a Security Agent: Models, Harnesses and Evals
Can an AI security reviewer find vulnerabilities it has never seen—or is it simply repeating familiar benchmarks? In this three-hour hands-on workshop, you’ll build a security-review agent and the evaluations needed to…
BAML by Example: Build a Self-Improving Agent
Give an agent a goal in plain language. In this hands-on workshop, you’ll use BAML to build an agent that creates the tools it needs on demand, critiques its own failures, rewrites those tools and improves across…
AI-Powered ESP32 Development with PlatformIO
AI-assisted development is reaching every part of software engineering—including embedded development on microcontrollers such as the ESP32. In this introductory, hands-on workshop, participants will connect an AI…
Context Engineering: Building Better Context Systems with DSPy
Context systems often fail in ways that better prompting alone will not fix. In this hands-on workshop with the author of O'Reilly's The Context Engineering Handbook, you’ll diagnose a range of common—and less…
Moving Your Agentic Coding Workflow to Open-Weight Models
Claude Code, Codex and other AI coding agents are transforming how developers work. But their cost, privacy implications and dependence on proprietary models are leading many teams to explore more open alternatives.…
Building Agents That Safely Operate Real Cloud Infrastructure
AI agents can inspect and change cloud infrastructure—but how do you let them act without giving up control?
In this hands-on workshop, you’ll first build a read-only inventory and audit agent that queries live…
Beyond the Context Window: Building Shared Agent Memory
Your coding agent learns how a project works, then the next session repeats the same investigation. Another agent misses an architectural decision, follows an outdated instruction or cannot access knowledge held in a…
Building for dataflow: the developer stack behind fast inference
An interactive, half-day session focused on SambaNova's dataflow developer tools, detailing how engineers deploy and fine-tune models designed for Reconfigurable Dataflow Units (RDUs).
You will compile a PyTorch model…