we’re a day late again, sorry guys.

No joke I pretty much write all the newsletters by hand and it takes me about 6 hours each week.

pls read the builders notes this week ive got some sauce for you.

Quick one before we get into it. My good friend Ashton is hiring at Intraflow.

They build and deploy custom AI agents/AI Workforces for established businesses.

Python agents on Docker + AWS, plugged straight into Gmail, Xero, Salesforce, GoHighLevel, Slack. running autonomously inside real companies, every day.

they're looking for some absolute weapons to join the team. AI Agent Engineers with:

• Production Python - 2+ years, clean and maintainable code

• Shipped real AI agents on Claude or OpenAI APIs in production (not demos)

• AI-native: builds daily with Claude Code and Codex

• REST API + OAuth 2.0 experience (Gmail, Xero, Slack, Salesforce, or similar)

• Docker for containerising and deploying to prod

• Builds client-facing dashboards in low-code tools like Retool

• Owns production systems end to end

• Thinks in agents and autonomous systems

remote to start. they want someone in it for the long haul, with a real path to a director seat as Intraflow scales (and they are scaling fast).

if that's you → apply here: https://form.typeform.com/to/dllbWRsj

if you know someone who fits, flick their name to admin@intraflow.com.au with subject line “Referral — [Name]". Ashton's paying $500 AUD for every referral they hire.

📌 TL;DR

• Notion's Developer Platform → your workspace is becoming a shared canvas where humans, agents, live data, and context all work together.

• Claude for Small Business → Anthropic is packaging Claude as an AI employee for finance, sales, marketing, support, HR, and ops.

• Claude Agent View → managing 3, 5, or 10 Claude Code sessions is becoming more like running an agent control room.

• Codex /goal → OpenAI beat Claude Code to a genuinely useful long-running agent feature, and Hermes copied it fast.

• Builder's notes → i’m splitting my AI setup into a personal AI OS and portable business agents that can move between tools.

Notion might of solved the biggest issue with AI and teams

notion have absolutely cooked.

This right here ladies and gentlemen is the future of agent and team collaboration.

That link takes you to the exact timestamp. Even if you don't use Notion, watch it.

They launched their Developer Platform, and the gist is simple: your workspace becomes a shared canvas where you, your team, and your agents all work on the same project.

Shared context has been the biggest unsolved problem in AI.

Anyone who has tried running Claude Code with their team knows what i mean. Your team is in one place, your agents are in another, and context is scattered everywhere.

Briefs in Notion, customers in Stripe, ad data in Northbeam, sales notes in HubSpot. Then Claude Code grinding away on your laptop, invisible to everyone else.

absolute nightmare.

Notion shipped three things to fix it.

1. External agents. Claude, Codex, openclaw, hermes, or whatever you've built. They all plug into Notion like collaborators. Tag them into tasks, ask questions, push work forward on the same page your team is already using.

2. Live data syncs. Notion pulls live data from your other tools, so your agents aren't working off stale data.

3. The Notion CLI. Gives Claude Code, Codex, and other local agents a better way to use Notion. They can read pages, update docs, and edit databases much easier than with the notion MCP

The demo video is a dev workflow, but the same pattern fits anything.

Say you're launching a new product.... the brief lives in Notion. Claude drafts the landing copy. Your marketer looks over it and approves. Codex builds the page. Your designer wants to chime in and make a change.

Agents, humans and context all in one place working together

I think this is a big deal - I've never seen a compelling solution like this before.

OpenAI beat Anthropic to a genuinely useful agent feature

I know. Weird sentence. OpenAI released a cool agent feature first. Someone check on Anthropic.

OpenAI shipped /goal before claude did.

/goal is mainly built for coding and development tasks, but the idea applies anywhere you have a big, ambitious job with a clear finish line.

One user on X said he had Codex running for 50 hours on a single goal.

That is the point.

For example: "Find 100 strong candidates for this role. Only include people with proof they’ve done the job before, a relevant portfolio or work history, and a way to contact them. Write a 2-line reason for each candidate and keep replacing weak matches until all 100 are worth reviewing."

Or: "take my Shopify store from a 42 mobile PageSpeed score to 90+, without changing how the site looks. Keep testing, compressing, removing junk, and re-testing until the score is green."

Here’s the cleanest way to think about it.

Normal agent work still feels like managing a worker between attempts. You check the output, spot the weak bits, remind it what you wanted, and send it back in.

With /goal, agents manage more of the project themselves.

It keeps a kind of project notebook: what it tried, what still looks broken, what needs checking, and what "done" actually means.

Codex added this publicly in on April 30. Hermes Agent copied the pattern the next day and explicitly credited Codex, which i love. Good idea ships somewhere, everyone steals it, the whole category gets better.

Claude Code added its own /goal on May 11.

Claude is getting an agent control room

It wouldn't be a normal newsletter without new Claude features...

First, they launched Claude for Small Business: 15 skills and plugins for finance, sales, marketing, HR, customer support, and ops.

It plugs into tools like QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace, and Microsoft 365.

So you can run things like invoice follow-up, payroll planning, month-end close, lead triage, cash-flow analysis, campaign creation, and support workflows inside Claude Cowork.

If you’re a small busines owner just getting your feet wet with claude cowrok - this is pretty cool :)

The second update is Agent View, pulling multiple Claude Code sessions into one view.

It's a proper control room for managing all your claude sessions in one place.

Anyone who uses claude code regularly knows how messy it gets managing 3, 5, 10 agents all running at once.

I think this would pair really nicely with the new /goal feature - to manage your long running agents.

fyi this agent view is only visible in the terminal/ CLI

Also this week...

• Perplexity Personal Computer is now available to all Mac users → Perplexity’s Mac app can now work across local files, native apps, the web, and 400+ connectors.

💡 Builder’s notes

I've been building a lot with Hermes Agent this week, and i think it's helped me clarify how i want my AI setup to work.

hermes agent is basically the new kid on the block that has replaced openclaw

Basically, i think everyone needs 2 layers of AI.

1. A personal AI OS. This is the layer you sit in all day. For me, that's still Claude Code. Email, writing, building websites/ apps, research, marketing campaigns, random business tasks. Claude is my daily workhorse.

2. Business AI infrastructure. These are the agents that run in the background and own a specific function. Client onboarding, customer support, lead follow-up, brand deals, reporting, that kind of thing.

Claude is brilliant for the first layer.

i'm less convinced it's the right home for the second one.

The agent i've been building this week is a brand deals manager.

It checks my emails on a schedule, finds sponsor opportunities, triages them, tracks negotiations, drafts replies, follows up, and reports back to me in Telegram.

I’ve setup a full workforce of agents I can chat to in telegram…

Trying to build out this kind of thing feels awkward in Claude Code.

Scheduled tasks seem to break a lot for me, and the Claude Telegram integration kinda sucks, and anything running on my laptop dies the second i'm on a flight with the lid shut.

Hermes feels cleaner for role-based company agents because it's open source, model-agnostic, and designed to run in the background.

The model freedom matters a lot too.

I really don’t like the idea of being too beholden to one company, like I would be if I used something like Claude Managed Agents for this.

With Hermes, i can run Claude, GPT-5.5, local models, or whatever makes sense for the job.

I've been using GPT-5.5 a lot this week and it's surprised me.

GPT 5.5 now powers all of my Hermes agents through the $100/month Codex subscription.

I really like it - I think it’s a great daily driver model

The writing is finally much less ChatGPT-smelling, and the coding is excellent. but it does have this weird obsession with Goblins?!

Opus 4.7 still absolutely clears when it comes to planning, creative thinking, strategy and longer form writing - imo.

🧰 Tools to try

• Obsidian Headless → Very handy if you're trying to run agents on a server with Obsidian as the shared brain.
  

• Printing Press → CLI factory + library for agents. The pitch is simple: MCPs and official APIs often waste tokens, so this gives Claude Code, Codex, OpenClaw, and Hermes fast local CLIs for 30+ services, plus a way to print new ones.
  

• Negotiation Skill → i gave this to my brand deals manager agent and so far it's doing a good job. Higgsfield deal secured, Stripe is in the works.

🥣 Brain food

• brain.am, my focus playlist → WARNING: use at your own risk. You won't be able to stop working.

• How I automated Meta ads with Claude lol → my latest YouTube video. I walk through the exact workflow from last week's Builder's Notes, Claude building ads, writing copy, generating creatives, and setting up the campaign.