would you let an ai agent buy for you?

Hola amigos,

What a week it’s been baby.

📌 TLDR:

• Tracking skills for the business - who’s using them, etc?

• Testing different content formats on IG.

• AI moving beyond the chat window

• Owning your own AI

• And is marketing dead if everyone has an agent that is acting for them?

💼 1. What's new in the biz

A map of the skills we're building

I'm making a skills platform so people can track how skills are being made across the organisation.

Rough mockup being produced here which is cool

Once people start turning their work into reusable skills, you need a way to see what's already there. Otherwise someone spends an afternoon solving a problem that another person solved last week.

Plus what if one person upgrades a skill but then fails to share it properly?

And then how do you see which ones are actually being used?

The platform should make it easier to find a skill, understand what it's for and see who is working on it.

I like the idea of the organisation getting better every time someone figures something out. A useful process should be able to travel beyond the person who made it.

And we don’t necessarily want it stuck inside of ChatGPT or Claude - for what if someone wants to change later on)?

The content grind continues

I've also been trying to get more deliberate about testing different videos.

The frustrating thing I've been running into is that small changes don't seem to give me a clean new test on Instagram - it just errors out. To try another version, I end up making the reels much more different than I'd like.

If I change the hook, pacing and visuals at once, it's harder to know which change helped.

Instagram's trial reels are designed to show content to non-followers first. But they don’t let you really split test different elements very well - would be cool if they did.

I'd love a simpler way to change one thing, test it and learn something specific.

Content teams and systems are gate kept to people who have teams of 20+ and millions in revenue already.

It seems odd that there doesn’t exist something that allows me to turn my self-made content to the max and allow me to grow my brand, following and audience.

A related frustration: Zuck talks a lot about openness, while Instagram is pretty restrictive about automated access to its own content. Meta explicitly prohibits scraping without permission.

I understand the privacy argument. People shouldn't have their information copied and republished without their knowledge.

But also don’t advocate for openness all the time (as per his most recent essay)?

A rebrand

I feel like people constantly under index on brand design.

When you’re starting out it generally doesn’t really matter - since you’re just trying to prove you’re not going to die.

But once you find your footing, it should take centre stage much more.

Accordingly, I'm thinking about a rebrand for the website, so I've been collecting design inspiration.

I’ve been using https://logoinspo.com/ and X and saving anything I like to Sublime.

I want to get clearer on how the business should look and feel, alongside how we explain what we do.

If there's a website, designer or resource you think is particularly good, send it through. I'd love to see it.

🤖 2. What's new in AI

Haiku gets cheaper

Anthropic released Claude Haiku 5.5 on 7 October, aimed at short, repetitive and high-volume work.

Standard API pricing is US$0.10 per million input tokens and US$0.50 per million output tokens for prompts up to 100,000 tokens. Above that threshold, it's US$0.50 and US$2.50. Anthropic also halved Sonnet 5.5's cache-read price. The announcement

What does this mean?

Another day - another model. To be honest I’ve just been using Opus 5.5 and Sol.

But if you’re worried about what to do

Sorting enquiries, extracting details from documents, checking records, summarising what changed. These are good places to compare a smaller model with whatever you're already using.

You still need to check the results. The cheapest model isn't necessarily the cheapest workflow if someone has to fix everything it produces.

The interface is changing faster

I think we're moving towards AI that fits into the day without requiring you to keep opening a chat window. I’ve talked about it before but this week there were a lot of announcements.

Glasses, rings, voice interfaces. Maybe a necklace or chain becomes another way in.

But the point is we will move away from a future where we are so reliant on screens.

View @carloAI’s post on X

The form matters because it changes when you can use the technology. Having an idea while walking is different from sitting down at your laptop to write it out. Asking about something you're looking at is different from taking a photo, uploading it and explaining the context.

Meta also had an awesome product launch too.

They’d launched these at Meta Connect but this is a really neat way to show off the new tech.

View @jimmyinvest’s post on X

These all point towards a more natural relationship with the technology. Less effort getting the information in, less time navigating an interface.

My instinct is that, for a useful assistant with controls I trust, the benefits of sharing more context will outweigh the privacy cost.

An assistant that knows your preferences, commitments and what you're working on can be much more useful than one you have to brief from scratch every time.

That doesn't mean giving every company everything. The trade-off depends on who can access the information, what they use it for and whether you can take it back.

We’re already seeing this with Instinct - people absolutely love it (which honestly surprises me a bit)

I think a lot of people will accept that trade when the benefit is concrete enough. Whether the device belongs on your finger, face or desk probably matters less than whether it actually helps.

A new personal (AI) computer is in town

Ghost AI's Core is coming for the personal computing market.

The company describes it as a personal AI computer that runs models on the device, connects to your apps and files, and builds context over time. It advertises continuous operation without a subscription.

Honestly this is so sick.

The founder is a 19 year old and has already raised a bunch of money.

View @zainmfj’s post on X

I think they tackle an interesting question to the ownership question: how much of your assistant should live on hardware you control?

What I love about this is the vision it is painting.

One where AI just ‘gets you’.

Your fridge is low on orange juice?

Your local AI has already scanned five different shops, found the cheapest price and sent you a message to place the order.

Saw on your whoop that you were low on sleep? It’s already made you a second coffee for the day.

We are moving into a future where AI just gets you.

And where AI may likely know you better than yourself and will be able to help you move towards this accordingly.

Software is dying again?

There's been a lot of discussion recently about software being dead.

Well, this week is no different.

It’s increasingly becoming the norm that it’s possible to basically decompile any proprietary software and then use it as you see fit.

With apps, let’s say Adobe Photoshop, the app is compiled from human readable code into binary code - which is basically a bunch of 0s and 1s. So to decompile that software into an actual app would take years (and someone really good at translating between the two languages).

Well now with AI you can point your latest LLM at the model and then just have it translate from the machine code (0s and 1s) into human readable code and then get your app rebuilt.

And voila, you have your own perfect Adobe Photoshop.

View @firesidealpha’s post on X

Any software that will continue to have some semblance of a business/moat will be anything that is server-based or relies on multiplayer.

So expect to see this moving forward.

Otherwise I expect there to be limited growth moving forward for these companies.

3. New tools and things I'm building

A content platform

I've made an editing skill and have been working on tracking trending content.

The idea is to make more of the content process repeatable. Find an interesting angle, work out what makes it worth covering, shape the piece and bring it closer to the style I want.

The bottleneck is not necessarily interesting ideas and things to talk about. The bottleneck is packaging them in a way that is digestible.

Tracking trends gives me material to consider. The editing skill helps me do something useful with it.

I still want the final piece to sound like me. A repeatable process should make it easier to express an idea, while leaving me responsible for whether the idea is worth expressing.

The content tracker fleshed out

On that point - this is the tracker in a bit more detail.

Claude Code pulled every reel five creators posted over the last 30 days: 185 reels and 54.7 million plays. It flagged the 23 that did at least three times their creator’s normal views. That took under two minutes and cost less than a dollar.

Then I handed every reel to Jev, a model from TypeSafe AI that doesn’t write anything, it only judges. You give it a post and a set of typed questions (what’s the format, what’s the hook, how easy is this to remix for me) and it gives back answers with confidence scores. It made 1,480 judgments in 2.2 seconds for less than a cent.

Was super curious about how one could use Jev so figured this could be a useful way to track it.

Instagram post

It’s cool - but very difficult to explain to non-technical people.

The easiest way was to hook it into Claude/Chat as a plugin and then have Claude call on it when it needs it and actually never think about it again.

The trick is to compare each reel against its own creator’s median, not raw views. Someone with five million followers will always beat someone with fifty thousand, so raw views tell you nothing. Comparing a reel against its own account shows you which ideas actually broke out.

A caveat: we skipped trial reels. They get tiny view counts and drag the median through the floor. On my first run, one creator’s median came out at 3,000 plays. Once I skipped trial reels it was 133,000.

If you want to build your own, I wrote up the full setup with the prompts: The Competitor Content Tracker.

Meeting context into an interactive deck

I've also made a deck skill that pulls context from our meetings and out WhiteHorse AI Brain, matches it with the brand design and produces an interactive deck.

There is a lot of useful material in a meeting that can get lost between the conversation and the presentation. What someone is trying to achieve, the examples they gave, the questions they asked.

What’s super cool is that it will actually pull exact quotes from the meetings so that the exact pain points of people are quoted in the deck.

Obviously I need to check the output but its been pretty fkn cool.

But it's a beautiful example of the kind of workflow I want to keep turning into skills (and hopefully for you too).

Take the context we've already gathered and carry it through to something we can use.

🤔 4. Differentiated take: Does the best product start winning more often if agents are doing all the buying?

Imagine a customer who has the time to compare every relevant option before buying.

They understand the fees. They read the conditions. They remember to cancel the subscription they stopped using. They can explain exactly what they want and keep looking until they find something close to it.

With the rise of agent commerce, it feels like we're putting a PhD-level research assistant in everyone's hand. The answers won't always be right, but the amount of effort someone can bring to an ordinary decision could increase substantially.

Patrick Collison posted some thoughts on agents in the economy this week. One of his predictions is that agents will make product quality a more rewarding strategy, as consumers become better equipped to evaluate what they're buying.

And I think that will be increasingly true..

Think about the small things people put off because they aren't worth the hassle in that moment.

Comparing a recurring bill. Working out how much money needs to stay available for upcoming expenses and what could move into savings. Finding the subscription you no longer use, then getting through the cancellation process.

Each decision might be manageable. Doing all of them consistently is another job on top of everything else in your life.

An agent will make that work easier. It could notice, compare and bring you a decision, rather than relying on you to remember that the decision needs making.

That could be good for consumers and uncomfortable for businesses that benefit from people doing nothing.

There are more complicated possibilities too. An assistant could help organise financial information or flag a question to take to an accountant.

The broader point is that more decisions will get the attention they deserve.

And there will actually be a more efficient market that is formed.

Your agent doesn’t really care about the ads on Amazon.

It can detect the bot farm on a twitter post.

That doesn’t mean traditional marketing is gone but it is interesting.

Does marketing become less important?

I think some kinds might. If a product mainly wins because it is the first one someone sees, a patient assistant could make that advantage less reliable.

But a good product still needs to explain itself. Buyers need to understand what it does, who it's for and why it fits their situation. Evidence, reputation and clear information could become even more useful.

And there's no single best product for everyone. One customer wants the cheapest option. Another cares about support, design, familiarity or how quickly it arrives. An agent needs to understand those preferences before it can compare anything sensibly.

Then there's another question: who does the agent work for?

Does it search for the best outcome for me? Or can a business pay to become the recommended option?

If the recommendation is influenced by a commercial arrangement I can't see, I still have to work out whether I trust the person making the recommendation. Except now that person is software.

It really is a bit of a black box in terms of what happens under the hood when agents go and search for you.

But I do think there is a good outcome: better buying decisions could give better products more chances to win.

And produce a more efficient rational market for both buyers and sellers.

If you knew your next customer would arrive with an assistant that had done PHD level homework, what would you change about your business?

5. Carve-outs

A visual piece worth a look: this post from LegalPrimes, plays with the idea of seeing movement across time.

Olivia Dean is back in Aus and is crushing it.

Peace,

Alex