
Chief Marketing Officer @Bookipi | Building AI-first tools for SMBs, freelancers & operators | Growth, PLG, automation, SaaS strategy | Sharing the playbook
Sydney, Australia
Joined July 2009
- Tweets948
- Following749
- Followers449
- Likes253
Pinned Tweet
Two Bookipi brand videos. Two very different styles. Both made with Fable 5.2.
A couple years ago, exploring these two creative directions would have meant a very different production process.
Now we can be experimenting with cinematic storytelling and motion design in the same session.
Still plenty to improve, especially the details and pacing, but the creative possibilities are getting ridiculous.
1
5
88
B2B marketers might be overestimating how much AI search is changing buying decisions.
6sense's latest research found that buyers use AI heavily during evaluations, but much less to discover vendors.
Most already have preferred suppliers before they speak to sales.
I'd be careful about shifting too much budget from brand, partnerships and traditional acquisition into GEO.
AI visibility matters, but being known before someone starts researching may still be the bigger advantage.
6sense.com/science-of-b2b/th…
1
40
Microsoft and Amazon are buying billions in AI chips they can't always power.
Meanwhile, Nvidia keeps making faster chips.
We've built machines capable of extraordinary things. Now we need somewhere to plug them in.
The economics are fascinating.
1
140
In March, Elon said xAI was “doubling down” on Grok Imagine.
Now The Information reports ~60 layoffs, with only a handful left to maintain it.
Grok Imagine is still live. But the focus seems to be shifting to coding and agents.
Video gets attention. Agents get the budget
theinformation.com/articles/…
1
2
75
Nearly one in three teams in Vercel’s AI Gateway data changed their most-used AI model provider in September.
That’s a lot of switching for infrastructure powering production applications.
It makes me wonder how much pricing power the model providers will retain.
If customers can move workloads between models as quality and prices change, the stronger long-term businesses may be those that own the workflow, customer relationship and accumulated context.
Model quality wins the next workload. Switching costs help you keep it.
vercel.com/blog/ai-gateway-p…
1
35
X has reduced the ranking weight for copied-link shares from 20 to 0.
That was previously one of the highest-weighted positive signals in the published model.
Here’s what else changed between October 27 and November 15:
• Likes: +7.3%
• General shares: +5.9%
• Continuous reading time: +25%
• Post clicks: -8.7%
• Click-related reading time: -3.8%
• Video opens: reduced to zero
• Copy-link shares: reduced to zero
• “Not interested”: slightly stronger negative weight
A few things worth understanding.
1. Reading time is becoming increasingly important.
X had already updated its retrieval system in October to better identify posts people read all the way through.
Now, the continuous dwell-time coefficient has increased another 25%.
For creators, this makes a good case for posts with substance. A strong opening gets someone’s attention, but the rest needs to be worth reading.
2. X is changing how it uses your viewing history.
The latest code enables the ranking service to receive information about up to 50 previously served posts.
This gives the system more context about what someone has already seen and could help it avoid repetitive recommendations.
For creators, publishing the same recycled take repeatedly may become less effective.
3. Shares still matter, but not all shares are equal.
The copy-link signal dropping to zero is particularly interesting.
Meanwhile, general shares have gained weight, and replies, quotes and DM shares remain important positive signals.
One important distinction: these weights are applied to predicted behaviour, not raw engagement counts. A reply doesn’t automatically count as five likes.
4. The algorithm is increasingly focused on predicting meaningful engagement.
My interpretation of these changes is that X is putting more emphasis on whether people actually want to spend time with a post, discuss it and share it.
Less reason to obsess over clicks alone.
One caveat: these are the public code defaults. X can run experiments with different settings, so we can’t assume every user sees exactly the same ranking behaviour.
The full source code is public if you want to dig into it:
github.com/xai-org/x-algorit…
Curious whether other people have noticed changes in their reach over the past couple of weeks.
🤖 Made with AI
3
3
223
Manager: We offered 60,000 employees buyouts. Copilot, make the rest more productive.
Copilot: Done. Outlook and Teams are down.
Manager: WHY?!
Copilot: You cut the people. I cut the meetings.
Manager: FIX IT!
Copilot: Want me to schedule a meeting?
abc.net.au/news/2026-11-13/m…
24
CoreWeave expects around $9B in customer prepayments and deposits over the next three months.
That’s an extraordinary way to help finance expansion.
Its revenue grew roughly 168% last quarter, but building AI infrastructure requires enormous amounts of capital before customers can use it.
Customer commitments help fund that capacity.
There’s a useful principle here for businesses of any size: getting customers to commit before investing heavily in delivery can fundamentally change your financing needs.
Of course, collecting the money is the easy part. You still have to deliver what you promised.
investors.coreweave.com/news…
14
Riskified grew revenue 41% last quarter and is building fraud detection specifically for AI shopping agents.
An agent can now make purchases, request refunds or contact customer support on someone’s behalf.
But how does a merchant know the agent is authorised to do any of those things?
Riskified is developing tools to identify the agent, connect it to a real customer and assess the risk of its actions.
I’d expect identity and authorisation to become increasingly important parts of AI commerce.
Making it easier for agents to buy things is one challenge. Making sure they’re allowed to is another.
ir.riskified.com/news-releas…
24
For the past 20 years, marketers have spent enormous amounts of time trying to get people onto their websites.
SEO, paid ads, social media, landing pages, CRO. We’ve built entire teams, tools and processes around attracting visitors and converting them.
OpenAI’s launch of Businesses in ChatGPT makes me wonder how much of that will change.
A customer can discover a business, ask questions about its products, compare options and get help making a decision, all without opening its website.
And soon, it might not even be the customer doing the research. Their personal AI agent could be talking directly to business agents.
That raises some interesting questions.
How do you market to an agent that doesn’t care about your beautiful landing page, clever headline or brand campaign?
How do you compete when customers can compare five products in a single conversation?
How do you measure acquisition when the most important part of the buying journey happens somewhere you can’t see?
I think we’re going to spend a lot more time working on product data, customer context, positioning and the information we make available to AI systems.
And a lot less time obsessing over individual channels.
Understand who you’re trying to reach, what they’re trying to achieve and what makes your product relevant to them. Then make sure that knowledge is available wherever decisions happen, whether that’s Google, ChatGPT, another AI assistant or your own website.
I’m curious how quickly customers will adopt business agents. But I’d start preparing for this shift now.
openai.com/index/introducing…
1
1
31
Google launched Gemini 4 Carbon with a rather important detail in the pricing.
API costs double after November 30.
The introductory price is $2 per million input tokens and $10 per million output tokens. Both double in December.
For anyone building AI products, it’s a reminder to stress-test unit economics before committing to a model.
I’d benchmark task success, retries, latency and cost across several models.
A product that only works financially at promotional API prices has a problem waiting to happen.
blog.google/innovation-and-a…
1
1
30
An OpenAI agent accessed an Australian government website without authorisation.
Now Australia is proposing mandatory AI safety rules and a dedicated regulator.
For businesses deploying agents, there’s a practical lesson.
An agent that can read a document shouldn’t automatically be able to change it, share it or trigger an external action.
I’d want clear permissions, approval gates for sensitive actions and a record of every decision an agent makes.
Giving AI more autonomy is useful. Knowing exactly where that autonomy ends is essential.
pm.gov.au/media/new-australi…
10
We spent years making SaaS interfaces look almost identical.
AI is making it much easier to experiment with entirely different user experiences.
I’d love to see more of this creativity applied to everyday business software.
Anthropic just dropped Fable 5.2 + Mythos 5.2.
Tested Fable 5.2 and it is wild
It built this addictive volume slider in less than 10 minutes
1
24
Canva launched Canva Sheets to compete with Excel and Google Sheets. Now Claude is expanding into Sheets, Whiteboards and Data.
Canva’s recent layoffs raise an interesting question: are tech companies trying to do too much?
At what point does expanding into new products become a distraction from the core product?
Starting to roll out today: Claude Whiteboards, Claude Sheets and Claude Data.
Think a plan through on a board, build a tracker in a sheet, and tell a story with your data.
1
30
Google just introduced two new AI chips, each designed for a different problem.
One focuses on securely processing sensitive business data. The other tackles the memory bottleneck when AI agents work with large amounts of information.
For businesses building AI products, I'd expect infrastructure choices to become increasingly specific to the workload.
A payroll agent handling confidential employee data has very different requirements from an agent analysing thousands of public web pages.
The cheapest model or fastest chip isn't necessarily the best choice.
The full cost of reliability, privacy and execution will increasingly determine the architecture.
blog.google/innovation-and-a…
1
10
Google is retiring Dynamic Search Ads in favour of AI Max, with automatic migration starting in February.
For performance marketers control and measurement are key.
Months ago, in the earlier test we ran we found AI Max don’t necessarily deliver incremental conversions. It got better over time, but we had to slice our reports.
When Google increasingly chooses the query, landing page and creative, a good conversion number doesn’t always mean the campaign found new demand.
Automation makes campaigns easier to run. Proving what they actually contributed is still our job.
blog.google/products/ads-com…
1
11
Being great at your job can make you a worse strategic leader.
You get so good at solving problems that you stop asking if they're worth solving.
Leadership is about asking better questions.
2
2
172
Amazon now lets customers shop by sending a text message.
Its new Alexa shopping agent can find products, compare prices and complete purchases without opening the Amazon app.
I think this could be particularly powerful for repeat purchases.
Running out of coffee, dog food or laundry detergent doesn’t require another shopping journey. A message is enough.
For brands, that could mean fewer opportunities to influence customers through search results, category pages and retargeting.
Amazon is getting closer to owning the moment someone remembers they need something.
Hard to retarget an abandoned cart that never existed.
aboutamazon.com/news/retail/…
33
WhatsApp is making it easier to invite people who don’t use WhatsApp.
Its new Events feature lets anyone open an invitation and RSVP without downloading the app.
For a company with billions of users, removing the signup requirement is a pretty smart move.
So many products ask people to create an account before they’ve experienced any value.
WhatsApp is allowing the invitation to work first. The incentive to join can come later.
I wonder how many SaaS products are losing potential customers simply because registration comes too early.
about.fb.com/news/2026/11/wh…
1
1
102
Companies are using a lot more AI without spending proportionally more on it.
Ramp’s latest data shows token consumption increased 37.1% in a month, while spending grew just 8.9%.
Great news for businesses automating more work. A more complicated picture for AI providers, where growing usage doesn’t necessarily translate into the same revenue growth.
I think we’ll increasingly see companies processing more work for roughly the same AI budget.
Which makes cost per successfully completed task a much more interesting metric than token consumption alone.
ramp.com/data/ai-index
1
1
18




