📩 it's a skill issue

📜 TLDR:

• Opening up limited spots for 1-1 AI consulting for founders

• Claude tries to watermark its output (the irony)

• Skills are all you need

• Anthony Bourdain and the purpose of life

Enjoy :)

💼 1. What’s new in the biz

I'm opening a small number of slots for one-on-one sessions for AI consulting with business owners.

The idea is simple. I sit down with a founder, we go through how they're actually using Claude or ChatGPT day to day, and we fix what's broken. Not a demo, not a pitch. Just working through the specific reasons it isn't sticking for them yet.

I keep having a version of this conversation anyway. Someone tells me they've tried AI properly, they got a few good answers, and then it drifted back to being a slightly faster Google. Nearly every time, the problem isn't the model.

It's structural.

They're starting from scratch in every conversation, or they've never given it the context that would make its answers actually about their business, or they're using it for the wrong half of their week.

Those problems are fixable in a couple of hours.

If you’re at all interested in what this looks like for you or your team, simply reply to this email and we’ll see if we can help.

🤖 2. What’s new in AI

Claude wants to put a watermark on its own output

Anthropic now embeds an invisible watermark and provenance data into everything Claude produces, worldwide.

The stated reason is reasonable enough: people should be able to tell what was machine-generated.

But think about how the product got built. Claude was trained on essentially the entire internet. Every blog post, every forum thread, every book, most of it scraped without asking anyone. They settled a class action with authors over exactly this for $1.5 billion in July.

So everything went in unmarked, unattributed, and largely without permission. And now what comes out gets a label.

I'm not saying watermarking is wrong. Provenance is probably good and I'd rather have it than not. Existing in a world where you don’t know what is machine generated and not is less than ideal.

It's the asymmetry that is quite ridiculous. The industry's position, roughly, is that tracing where the inputs came from was impractical and would have killed the product, but tracing the outputs is a moral obligation. Both of those can't be the principle.

One of them is just whichever rule happened to be convenient at the time.

Meta wants all your data

Meta released an open model called Muse Glimmer. Thirty billion parameters, Apache licensed.

Basically it means that you can (if you have a good enough computer) run it on your own laptop.

This comes after Zuck’s manifesto on ensuring AI should not be contained to “just a few”. (would recommend having a read of it)

The pitch is that everyone should get a personal superintelligence rather than it sitting inside a handful of labs. His line: "Everyone will have an exceptionally capable personal agent that understands you, your goals, and everything you care about."

But now read that with your meta glasses on.

This is the company that makes all of its money (literally 95% of their revenue still comes from straight ads) from knowing who you are and serving that information up to advertisers.

Having said that it is a picture worth aspiring to. Having an abundance of intelligence for everyone, is much better than it being caught in a few hands.

AI is showing up inside the software you already pay for.

The case for software.

Wherever your data is - these places have an in-built advantage. Slack, Salesforce, Clickup and more.

They have all your data, you know how to use it, there is a network effect to shared software (I’m better off if you also know how to use it).

What’s super cool is that they’re now shipping cool features, in a really digestible way inside of their software.

E.g. Slack shipped agents you can talk to in a DM. MYOB turned its AI BAS tool into a paid add-on. QuickBooks added auto-bookkeeping and a payments agent. Posthog added the ability to simply “tag” your agent. Lemma has added this too.

For a couple of years the assumption has been that cheap building would let small challengers eat incumbent software. Some of that is happening. But these three have bolted AI onto the advantage they always had, which is that they already hold your data and your team already has the app open.

I think this will only occur more and more moving forward. It’s super cool seeing how AI is now accessible to people in such simple ways.

Other quick hits

FlockOS. 

Flock Safety is an AI powered camera system. Valued at $8.4 billion company running number-plate cameras across thousands of American towns, feeding a national database police can search without a warrant.

Actual insanity ngl.

You’d think of all places, America, the land of the free, would be the last place for this to occur.

But what makes it legal is a 1970s Supreme Court ruling that says data handed to a third party loses its expectation of privacy. This obviously has some poor consequences…

The only control on searches is a text box where an officer types a reason.

An Idaho sheriff ran his wife's plate 700+ times in three months, reason given: "test." (yikes)

Fireship explains this further if you’re interested to know more.

Anthropic's IPO and a $2t valuation 

Anthropic is reportedly going public as early as October at around $2 trillion, which would be the largest listing in history.

And this valuation is not without cause.

What’s been so impressive is their growth.

They’re making real money. 

$74.1b ARR!

With an aggressive focus on enterprise sales, they have been seeing absurd growth.

Having said that the bear case for Anthropic is concrete: they'd need roughly $59 to $79 billion in annual profit to justify it, against revenue near $10 billion at the start of this year.

👷 3. Builder’s notes: Grokbot, Skills and 7 other tools that you can use right now

xAI released GrokBot

The Cursor acquisition has already released a number of bangers, most notably, GrokBot.

Twitter tweet

This is xAI’s all in one agent app, which lets you create agents for specific use cases, more formidable than skills, way less technical than setting up 6 different OpenClaw instances.

The thing I like about it is that it lives on its own computer. It runs whether or not my laptop is open, which sounds small but isn't.

The other thing is that it's the same in both places. The app on my desktop and the app on my phone are the same app. I don't build something at my desk and then find it missing when I'm out. That sounds obvious. But if you’re like me using Claude, Chat and 10 different other AI tools they’re often just so different when you go from mobile to desktop (which is really annoying).

The best use case?

You can use it as a chief of staff. It holds the daily stuff and fires off the tasks I'd otherwise carry around in my head.

Mostly what I appreciate is that it's simple, and that the parts you'd want to customise are the parts you can customise. That combination is rarer than you’d think.

The pricing is around $200/mo. Which may seem expensive.

But if you’re trying to use OpenClaw properly you have to connect it to a Codex subscription which is $150/mo anyway.

Plus doing that is quite complicated to set up - you need to know what a VPS is, what SSH is, etc.

Grokbot, I believe, provides a really easy way for non-technical people to get agents up and running.

Skills are the highest-leverage thing you can build right now

A lot of people are so worried about the next model. The next agent. The newest form of AI.

To be honest I think most of it is overkill. The easiest thing you can develop if you’re a business owner are skills.

A skill is a set of written instructions for a repeated job. How you onboard a client. How you write a proposal. How you triage the inbox on Monday. You write it once, properly, and then the AI does it that way every time instead of you re-explaining from scratch.

Three reasons I think they're worth the time.

They're just text. There's no platform to be locked into, no format to be deprecated, no vendor to go under. When a better model comes out next quarter your skills still work, because a better model reads text better than a worse one does. Most of what you could build right now carries some risk of being obsolete by Christmas. This doesn't.

They're a system of record for your SOPs. Most small businesses have their processes in one person's head. A skill is that knowledge written down in a form both a human and a machine can follow. You were told to document your processes years ago and didn't, because it was pure overhead with no payoff. Now there's a payoff.

They're yours. A skill is shaped around how you actually do the thing, not how software thinks you should. As long as you still run that process, the skill stays useful.

The compounding is what makes it worth starting now. Every skill you write is permanent and additive, so ten of them and your business genuinely runs differently, and none of that work is at risk from whatever gets announced next month.

7 other tools worth knowing

Everything here works today, and I've noted what plan you need - note most of these are gated.

1/ Pull your Google Drive straight into ChatGPT. Your Drive files now show up in ChatGPT's library, including things shared with you, so you can drop a doc into any conversation with an @mention instead of downloading and re-uploading it. Shipped 13 August. Needs Plus, Pro or Business, and the desktop app. This is the one I'd do first if you keep everything in Drive.

(Personally I have a model-agnostic setup, but if you’re just using ChatGPT right now could be helpful).

2/ Turn a spreadsheet into an actual app. Google Sheets got a canvas mode where you describe what you want in plain English ("a job tracker with status dropdowns") and it builds an interactive front end that syncs both ways with the sheet. Rolled out from 10 August. Needs Google AI Pro or Ultra, or Workspace Business Standard and up. If you've got a spreadsheet that half your team is scared to touch, this is the fix.

Alternatively if you’re just using Claude or Chat - you can deploy these as “artifacts” or “sites” respectively.

3/ Your meeting notes will now include the slides. Google Meet's "take notes for me" now screenshots what was presented and drops the images into the notes doc, instead of describing a chart in words. (Pretty cool feature)

4/ Claude now works inside your browser tabs. Cowork moved into the Chrome extension's side panel, so skills and connectors run against whatever page you're on and the conversation follows you to desktop and mobile. Shipped 12 August. Max and Team have it now, Pro is rolling out over the next few weeks.

I think we’ll see more and more of this streamlining - like Grokbot with xAI. Having a more streamlined way of doing things across the apps will become common place - assume Chat will do something similar.

5/ Ask Slackbot for the deck. Slack can now build a Google Slides or PowerPoint deck from your own template out of a thread, or a filterable dashboard, without leaving the conversation. Agents can also DM you directly. Needs Business+ or Enterprise for the deck and dashboard parts; the DM agents are on every workspace.

Very similar to Viktor - assume they saw the popularity of it and just said fk it let’s do this ourselves as well.

6/ Tell your team ChatGPT's free tier stopped rate-limiting text chats. Unlimited text conversations on the free plan since 6 August. Worth knowing if you've got staff who bounced off it months ago when they hit a wall, or if you've been paying for seats people barely use.

7/ Clean up your Shopify product data. Shopify products now surface inside ChatGPT, Perplexity and Copilot answers when someone asks about a category. Which means your product titles, descriptions and stock fields are now read directly by assistants, and badly formatted listings get skipped. Tidying them up costs nothing and is extremely helpful. For my e-comm friends out there.

All of these are super cool additions and are just each company’s attempt to deepen the moat inside of their app.

Let me know if any of these are helpful :)

🤔 4. Differentiated take: Anthony Bourdain and the meaning of life.

I talk with a lot of business owners (and people more generally) that are constantly worried about AI. I think it largely comes from a lack of uncertainty. The way I look at it though is that there are essentially two paths that could occur:

1/ we achieve the “singularity” by 2027, in which AI recursively learns on itself, we reach an intelligence explosion or superintelligence, in which intelligence is abundant (provided energy and compute are not constraints) and every problem ever is solved. If that’s the case, we’re all cooked and you needn’t worry anyway.

2/ AI continues to evolve, but we don’t reach said singularity. Instead it continues to improve, it dissolves into the economy via a long time-lag. It is a (very) useful tool that can serve our needs, and there will be opportunities to do things we never thought possible in our lifetime. 

I believe option two is more likely (I’ll explain further in a separate edition)

And so all you can do is focus on the things that are currently at hand and do your best with the information you are given. 

Having said all of that it led me down a bit of a rabbit hole. If we do achieve option 2, what then? 

Why do we do the things that we do?

I generally judge my life in order of the progress I am achieving in the different facets that are important to me. 

1/ Physical health (basketball, gym, sports more generally)

2/ Intellectual (what I’m learning on a weekly basis)

3/ Social (family, friends, relationships)

4/ Mental/emotional (meditation, etc)

5/ Business (growth in my businesses)

6/ Financial (pretty self-explanatory)

7/ Spiritual (deepening my understanding of my place in the world)

More recently however, I’ve come to appreciate the non-linearity of life. While this framework is still important and serves as a useful matrix for me, I’ve been thinking more about the essence of humanity. 

If we do achieve the “singularity” (which again I think is more unlikely than not but still an interesting thought experiment), then what? 

What does one do all day?

Which leads me to Anthony Bourdain.

(Yes I’m aware this

If you don't know him: he wasn't a celebrity chef. He was an unknown one. He spent years cooking in New York kitchens, most of them not good, before ending up running the line at Les Halles, a brasserie in Manhattan. He was in his forties. He'd been a heroin addict. He had no career to speak of and no reason to expect one.

Then at 42 he wrote an article about what really goes on in restaurant kitchens and sent it to the New Yorker. He heard nothing for months. Then one day an editor rang and said they wanted to run it.

What came after was Kitchen Confidential and then twenty years of going around the world eating with strangers. And somewhere in there he stopped being a chef who wrote and became something else. An icon, and a fairly odd one, because what people actually responded to wasn't the food.

It was that he wouldn't perform. He said what he thought because he thought it, and he'd sit down with anyone, anywhere, and take them seriously. In an industry built on polish he refused to be polished.

His show (and his ethos by the end of it) can be surmised as "Travel is at worst an opening of the mind and at best a lesson in empathy"

If you don’t know who he is, watch the above - you feel grateful to witness such a fascinating conversation.

So often in building a company around technology, saving time, increasing efficiency… I wonder what are we doing it all for?

Consuming Bourdain reminded me a lot of this quote from Dead Poet’s Society (RIP Robin Williams),

In short: that all the things we spend our time on are simply a means to enjoy the artistic end.

I think Bourdain presents an interesting perspective on life.

For him - optimisation of the human experience was: to do more random things, be exposed to new cultures, say yes to more uncomfortable experiences, and that in this short time in life you should aim to experience as much as possible. 

To put yourself in uncomfortable situations.

Because, fear is the tax you pay for everything worth doing. 

No human experience is original. It’s just new to you.

"Travel is fatal to prejudice, bigotry, and narrow-mindedness, and many of our people need it sorely on these accounts.  Broad, wholesome, charitable views of men and things cannot be acquired by vegetating in one little corner of the earth all one's lifetime." - Mark Twain

For someone who has aimed for so much linear progress in my life, opening my eyes to someone who experienced so much and ultimately concluded the best thing in life was to travel more and do more random shit is a powerful lesson.

It’s left me with the following: 

1/ Optimise for interestingness.

2/ Increase your surface area for being a part of something great.

3/ You can never control the outcome but you can give yourself the best chance possible to do amazing things.

🎶 5. Carve outs:

• Paul Graham’s essays (been told to read these by many people, just getting round to them now)

  • https://www.paulgraham.com/articles.html

• My friend’s substack: Everything is a Sales Funnel (I was going to write about this and he had already described it very aptly)

  • https://ameywrites.substack.com/p/everything-is-a-sales-funnel

• Michael and Dalton on YC (great if you’re into startups):

  • https://www.youtube.com/watch?v=pQErhgTsCOw

• - Anthony Bourdain: the Cajun Country episode (one of the best)

  • https://www.youtube.com/watch?v=-fPRzXD257w

• Basketball analysis by J Kyle Mann (one of the best analysts out there):

  • https://www.youtube.com/watch?v=tnzSx80SFfs

• LeBron starts his YT channel (actually crazy how many already famous people are cultivating their personal brands):

  • https://www.youtube.com/watch?v=e6BElTIS8x4

Otherwise,

Adios folks,

Alex