The Good
New CSIRO research pushes back on the “AI kills jobs” narrative. Analysing job ads from over 4,000 Australian firms between 2020 and 2023, researchers found that companies adopting AI posted 36 per cent more non-AI job ads over time than non-adopting firms – and were asking for broader skill sets, not narrower ones. The study argues that the real divide isn't between humans and machines – it's between firms embracing AI and those that aren't, with workers at non-adopting firms potentially falling behind their peers.
Comment: It's a refreshing counter to the narrative dominating global mainstream AI coverage – and it's distinctly Australian. The data suggests that the transition to an AI-powered economy doesn't have to be as disruptive or doomsday as the headlines imply, provided organisations (and their leaders) are thoughtful and intentional about how they bring AI in.

The Bad
In an intriguing article on frontier AI models and their impact on cyber security, the Australian Signals Directorate has highlighted the risks associated with AI becoming exceptionally proficient at coding. Whilst this enables businesses to do more and to do so more quickly, it also drastically lowers the ‘barrier to entry’ for bad actors to find and exploit software vulnerabilities. Both attackers and defenders in cybersecurity are getting stronger in the AI age, but it can be a ‘one step forward, two steps back’ situation if attackers can more easily find ways to exploit software. The article covers the best mitigation strategies for organisations to implement to ensure digital safety in the current era.

& The Ugly
In San Francisco, a 20-year-old man threw a Molotov cocktail at the home of OpenAI CEO Sam Altman, and made threats at OpenAI's headquarters. The man in question, Daniel Moreno-Gama, has been arrested and charged with attempted murder. Thankfully no one was hurt, but the event marks a disturbing step of violence amid increasingly polarising views of AI and the companies that are developing it. Altman took to a blog post to humanise himself – sharing a family photo and using the moment to reflect on his AI beliefs, personal mistakes, and call for de-escalation of rhetoric across the industry.
Comment: We believe there is plenty of room for criticism of AI and companies like OpenAI, Anthropic, Microsoft, Google, etc. but the contest of ideas should always be respectful, non-violent and constructive.

Learning AI shouldn't feel like drinking from a firehose. That’s why we’re building ClearAI Gym. Imagine a new online learning platform where you can work through short, practical classes at your own pace, built around the tools you're actually using and the context you're actually working in.

Think of it like a gym membership for your AI skills. You choose your class – ChatGPT, Claude, responsible AI, AI in marketing, AI in data analysis – and you show up when it suits you.

We built it because most AI training is either too technical or too generic.
ClearAI Gym is neither. But we would love to hear from you before we open the doors. Vote in our poll below.

I feel behind on agents. There, I said it. 

I run an AI consulting firm, have helped thousands of people learn the skills of AI and implement new solutions on a weekly basis. But right now, I feel behind. 
The pressure, and hype, around agents is real. Take this example: the CEO of Perplexity retweeted one user who spent $20,000 in four weeks on Perplexity Computer to "completely automate his business." The post got 130K views.

You see a post like that and think, “should I be doing that?” Every day there's another one.
But offline, things aren’t what they seem. A month ago I slacked Ryan Drake, who is neck deep in AI, about our progress with agents. The honest truth was that the agents we were running didn’t lead to meaningful change. This week, a friend shared the same frustration: they couldn’t even get OpenClaw into their Slack.

Whatever you see online, the reality is different (surprise). Knowing that doesn’t stop the build up of pressure. Honestly, I’ve been overwhelmed at the options. 

Should I use OpenClaw, Claude Code, agent swarms, Perplexity Computer, Hermes (not the luxury brand) or any number of new solutions?
Or should I try what I saw someone had done to build an outreach agent with some crazy response rate – hang on, that was SaaS, we’re consulting, will that work for us? Or should I focus on creating agents for our workflows? Or better yet, I’ve got two children, one on the way, and a growing business. I should optimise my health and focus, maybe I’ll create a Huberman coach. 
With hundreds of tabs, a dozen walkthroughs bookmarked, and five or so agents on life support, I was paralysed. 

And I don’t think I’m alone. If you’re reading this, I’d wager you’ve got your own version of this: a graveyard of half-started experiments, a bookmarks folder you’re embarrassed to open, and a growing sense that everyone else has figured out something you haven’t. Information bias has us running to gather data, rather than acting on what we’ve got. 

The solution lies in jam – the type you eat.
(Fun fact about this writer, I like to buy homemade jam on the road. My favourite was a marmalade purchased by the side of the road in Camargue last year.) 

In 2000, Sheena Iyengar ran an experiment in a grocery store. On some days, 24 types of jam were on display, on other days just six types of jam. The larger display attracted about 60% of passers by, while the smaller display attracted 40%. In theory, more attention leads to more sales. 
Of the people who stopped at the 6 jar display, 30% purchased a jar. Of those who stopped at the larger stall, only 3% purchased a jar. Ten times less conversion despite more attention. 

AI agents today are the 24 jar table on steroids. The opportunity is real, but so is the noise. And the noise is winning. 
The overwhelm you feel is caused by too much undifferentiated information. Every feed, thread, or demo adds another jar to the table and the most common outcome? You still have no agents running.

Some people are doing genuinely impressive work in the space. Garry Tan, President of Y Combinator has built an incredible suite of tools in GStack for coding, and now GBrain for OpenClaw. Andrej Karparthy’s home is entirely agent run – his OpenClaw (named Dobby) watches the cameras and notifies him when a delivery truck arrives. 

What these operators have in common is focus. They picked one application and went deep, not stopping at 80% (which is easy but useless), iterating instead to 98-99%. 
The gap opening up is between people who can focus and who can’t. That’s taste. Knowing what to build, when to stop, and what to ignore completely. 

The pattern looks like this: Every Monday you tell yourself this is the week you start building. You made little progress, opened up the last walkthrough you saw for inspiration, and by Wednesday you’re back to reading threads. 
In the time it takes to find and read those walkthroughs (let’s face it, there’s a lot of doomscrolling in-between), you could have built one agent, broken it, fixed it, and learned more than any tutorial could teach. The cost of paralysis is the compounding absence of learning by doing. 

Pick one thing. One agent, one workflow, one problem that you care about solving. Go deep on it. When you get to 80%, you’re halfway done. That’s when the real work begins. Don’t move on until you hit 99%.
It will probably feel like you’re falling behind while everyone is exploring new tools and you’re not. Maybe even put a social media embargo in place until you get your agent done. Here’s the bet: going deep on one agent will teach you more than hours of consumption ever could.

My next project is a second brain, borrowing from some of the work of Garry Tan and Andrej Karparthy to take as much of my personal history as I have – my message history, calls, contacts, emails, notes, reports, book highlights, podcasts and more – and pulling that together in a wiki my agent can access and update. 

That means I’ll be dropping my own social media consumption to get it done. I’ve got the tools (Claude Code + Codex) and the idea, so now it’s time to get underway. I’ll be reporting back with how I go – the good, the bad and the ugly.

I’d genuinely like to know, what are you building next?

Thanks for joining us. See you in the next drop.

Yours in humanity,