Featuring:

There’s a proposal you haven’t followed up on. An older article that still gets visitors but needs attention. A useful thing you’ve explained to customers twenty times and keep meaning to turn into something you can sell.

All perfectly reasonable jobs. All waiting for the same person.

Me.

That’s where my mind went reading about Astra.

OpenAI released GPT-6 Astra on September 3. Its pitch includes stronger computer use, better adherence to document templates, and an improved ability to keep track of the assignment when you make changes along the way. In its OSWorld 2.0 latency simulations, OpenAI reports higher performance with roughly 47% less time per task than GPT-5.6 Sol. Those are vendor test results, not a forecast for your business. OpenAI’s announcement

Still, that combination deserves a look. Especially if you work for yourself and the research, production, selling, and checking all land on your desk.

I’d start by reopening a few jobs you previously decided weren’t worth the effort.

Suppose you run a content business. An older article keeps attracting readers. You know it could use a better download, a clearer offer, or a follow-up email. But pulling the numbers, deciding what to change, and building the pieces never quite makes it onto the calendar.

Give Astra the article, a traffic report, the current offer, and examples of work you like. Ask it to identify one plausible missed opportunity and prepare the actual materials: revised copy, a useful download, and an email draft. Have it flag anything that requires new photography, testing, or facts you haven’t supplied.

Then check the work and publish the parts that hold up. Keep the final wording in your hands. Early reactions to Astra’s writing have been mixed, including criticism of how well it matches an established editorial voice. Decrypt’s review I’d test it on the research, assembly, and production work, then compare its copy with whatever already works for you. Your taste is still part of the product.

That gives you something specific to measure: whether existing readers become subscribers or customers more often. Another hundred visitors would be nice. Making the visitors you already have more valuable is a sensible experiment too.

For a consultant, the equivalent might be a service page that gets inquiries but leaves people confused about what they’re buying.

These are tests I’d run. Their value still has to be proved in your business.

Now take five open proposals.

Supply the original inquiries, your proposals, and the latest correspondence. Ask Astra to prepare a short review of each: what the customer wanted, what remains unanswered, and the next useful thing you could send.

Then have it draft the follow-ups using your actual scope and pricing.

I’d be looking for the unanswered question that makes a buyer hesitate. Perhaps the deliverable is vague. Perhaps nobody explained what happens in the first week. Perhaps your proposal somehow managed to spend three pages describing your process without telling the customer what they would receive. Yep, I have been guilty of that last one more than once!

AI can help assemble the evidence and prepare a clearer response. You still need to decide whether that response is appropriate.

Check every commitment before sending. An invented delivery date becomes your problem very quickly.

Count replies, conversations, and work won. The number of emails produced tells you very little.

This is the possibility I find most interesting.

You probably have expertise that would help someone with a smaller problem. A website review. A customer onboarding check. A practical assessment of why inquiries aren’t turning into bookings.

You’ve avoided offering it because the preparation and write-up eat the fee.

Consider a purely illustrative example. A $150 review takes three hours of your time. That’s $50 per owner-hour before other expenses. If you could deliver the same quality in one hour, including checking and corrections, with $10 in additional tool costs, you’d have $140 left for that hour before other expenses and taxes.

That doesn’t prove Astra can do it. It tells you what to test.

Use one past case. Give it your assessment criteria and an example of a finished report. Ask it to assemble the evidence and prepare the deliverable. Compare the result with your own work, including how much you had to fix.

If it holds up, you may have a small service worth offering to existing customers. Ask a few whether they would pay for it before building a whole new business around it.

The same thinking applies to a workshop, a planning kit, or a useful tool built from work you’ve already done. Your knowledge and customer relationships give the offer a reason to exist.

There is one number I’d write down during any of these experiments: how many minutes did I personally spend getting this into usable shape?

Include the briefing. Include finding the files. Include checking links and correcting the impressive-looking nonsense.

ARC Prize’s own testing is a useful reminder of why the setup matters. Astra’s best observed ARC-AGI-3 score was 62.7% with its standard setup and 99.9% with a provider adapter that preserves reasoning state. Those results used different reasoning settings too. You can’t turn the launch headline into a reliability promise for your workflow. ARC Prize’s evaluation

For your first test, supply a small set of files or exports. Direct work inside your apps depends on the tools and access available in the product you’re using. Define what a finished result looks like, and keep sending or publishing as a separate decision.

OpenAI says Astra is rolling out to paid ChatGPT plans within existing subscription allowances; additional usage can require purchased credits. API billing is separate. Check your account before buying anything else. Availability details

If the AI needs two hours of your supervision to save you an hour, that particular job hasn’t improved.

Tom’s Guide makes a useful suggestion: give Astra a whole job. Read the article

Here’s the business test I’d use. Attach an article that still gets traffic, its traffic report, and a sample of your best customer-facing writing. Then paste this:

Use the attached article and traffic report to build one useful email-signup offer for its existing readers. Choose the offer based on the audience and evidence in those files. Create the download, the signup copy, and a welcome email using my examples for tone. Flag unsupported claims. Save the materials as drafts; do not publish, send, or spend. If you lack a file or tool needed to produce them, tell me exactly what’s missing. Finish with the files, remaining decisions, and one metric to track.

You should have three drafts you can open and review. A checklist of suggestions means the job is still unfinished.

For example, a party-food article might become a printable shopping list and preparation schedule. A consultant’s onboarding article might become a client kickoff worksheet. Choose something your readers can use, and check that it actually works.

Write down how much of your time it took to get those drafts ready. If you publish the offer, track the share of article visitors who sign up. Sales may come later; an email address isn’t revenue.

Finish one experiment by Friday. If it saves you two hours, decide what those hours are for. Following up with a customer might bring in money. Taking the afternoon off might be exactly what you needed.

What’s the job you keep moving to next week?

Hit reply and tell me.

Workflows first. Tools second.

Affiliate disclosure: As an Amazon Associate I earn from qualifying purchases.

Take your last ten AI requests and ask the model what they would have cost at API rates. Most of us have no feel for this until we see the number. With frontier prices up 2.5x this month, it's a cheap way to find out which of your habits just got expensive. Use something like this prompt:

📝 75% of this newsletter is written personally by our team.

🤖 The remaining 25% is supported by AI—for things like topic research and grammar checks.

🦾 Each piece of content takes a lot of effort from our team, so your feedback means a lot to us. Let us know what you think!

To make sure you never miss an update, please move this email from the Promotions or Spam tab to your Primary Inbox. This helps ensure you receive the latest AI news, tips, and tutorials—right where you’ll see them first.

Reply to this email with any specific feedback or interesting insights!

Thanks for reading!