Using ChatGPT Agent Mode
August 4, 2025
Notes on branding, writing, and AI
How Far Will ChatGPT Agent Mode Get Me?
How about we wait for GPT-5.
I’ve been curious about tasks in Claude and Agent Mode in ChatGPT. Why not make a few tasks more efficient and put some agents to work on my behalf? Sounds awesome. But while it’s been interesting playing around with tasks and Agent Mode, the AI companies have a ways to go before this is more than just a novelty for most of us. Enterprise and dev users are further along (they have the resources to get past basic hurdles), but recent reports suggest that even for them most agents aren’t yet delivering on the agentic promise.
I connected Claude to my Canva account and really struggled to make anything happen. It certainly wasn’t as easy and intuitive as it should have been, given the excitement surrounding the integration. I also had Claude send some texts for me through Apple iMessage. I got it to work after a few tries, but Claude didn’t tell me that it was successful. In both cases—Canva designing and iMessage sending—it would have been much easier to just go to the other app and deal with the tasks on my own. So I’m still not understanding the point of those particular integrations.
What about Agent Mode in ChatGPT? I set up a seven step process to research topics, then write (using a style guide I put together for myself), fact-check, and edit a blog post three times a week. And then post it to my website as a draft for my edits and approval. I’d give the overall process and work a C+. Why?
First, it took forever for my Agent to think through all the steps and make a plan. I could have been halfway done with the first post by the time it had everything straight.
Second, it wasn’t able to log into my SquareSpace account using its proprietary browser—SquareSpace didn’t recognize it. When I asked my Agent to email the drafts to me, it asked for the credentials to sign into my email account. I told the Agent that it didn’t need to sign into my account to send ME an email. To that, it told me I could download the file and email it to myself. This is not helpful!
Third, the post itself was full of baggy copy, and it struggled to land a point. What’s more, though this is on the relatively trivial, head-scratching side, It went back and forth between British and American spellings of the word “color” for reasons that are not clear. (Should the project prompt mentioned it was a “pro proofreader”?)
Finally, when I had a test draft, I asked it (twice) to initiate the blogging project. Here’s the response: “Understood! I’ll begin the ongoing blogging project as planned. Starting next Monday…” Um, how about starting tomorrow? (Turns out that we trained AI well, which means it’s just as lazy as we are.)
What about the post? Was it any good? I’ll let you read it. I’ve made no edits. Leave your thoughts in the comments.
Here’s the post my ChatGPT Agent wrote:
Smart Mode, Smarter Business: AI Sneaks Into Your Everyday Tools
Here’s a confession: as someone who watches AI every week, I’m starting to notice that the most exciting developments are also the quietest. The headlines scream about model wars and “personal superintelligence,” but the updates that will actually change your workday show up inside tools you already use. Funny, isn’t it?
A smarter Copilot
Take Microsoft’s latest experiment. According to The Verge, the company is testing a new smart mode for Copilot that quietly chooses the best language model for whatever you’re doingtheverge.com. There are even hints of GPT‑5 hiding in the codetheverge.com. In internal descriptions, this mode promises to “think deeply or quickly based on the task”theverge.com. In other words, you no longer have to fiddle with drop‑down menus to decide if you need a research bot or a creative writer. The assistant just knows. (I mean, who needs rules, I guess.)
If that sounds like magic, marketing teams should take note. MarketingProfs suggests that smarter Copilot features could “enhance marketing automation and content generation” when GPT‑5 arrivesmarketingprofs.com. That matters because your competitors might be saving hours while you’re still copying and pasting between apps. Yet this news is also a reminder that the AI landscape can flip overnight. A recent analysis found that Anthropic’s Claude models now own 32 % of the enterprise LLM market, while OpenAI’s share has slipped to 21 %. Enterprises are choosing Claude because they value reliability and safety. Oh, and corporate budgets for AI doubled to $8.4 billion in the first half of 2025. Trust, not glitz, seems to be the new currency.
Generative design comes to Photoshop
If Copilot is making AI invisible, Adobe is doing the opposite: it’s making AI feel like magic in your hands. The company’s new Generative Upscale feature enlarges images to 8 megapixels and recovers lost detailtechcrunch.com. Harmonize makes a pasted object blend into a scene by matching colours, shadows and lighttechcrunch.com. And an updated object‑removal tool now focuses on just the thing you want gone, instead of smearing the entire phototechcrunch.com. As Adobe’s senior director Shambhavi Kadam notes, these features are designed to strip away friction and hand back your timetechcrunch.com. It’s the little stuff—those few minutes per image—that adds up when you’re juggling campaigns and invoices.
It isn’t happening in isolation. The U.S. Chamber of Commerce’s CO— platform notes that small businesses are already embracing AI writers like Anyword (which trains on your brand voice), content‑repurposing tools like Lately, and non‑GPT generators like Writeruschamber.com. Canva, Microsoft Designer and Vidyo.ai have added generative features of their ownuschamber.com. If you blink, you might miss the moment when your creative stack becomes an AI stack.
What does this mean for small businesses and writers?
First, remember that AI is no longer a luxury item. A FreshBooks survey cited by CO— found that roughly 25 % of small business owners were already experimenting with generative tools by mid‑2023, and two‑thirds planned to start soonuschamber.com. The reasons are practical: people are strapped for time and can’t afford extra headcount, so they lean on AI to repurpose copy, draft product descriptions and even design imageryuschamber.com. UK‑based Lambda Films used Midjourney to turn a director’s vision into client‑ready artuschamber.com. Solo marketer Anna Burgess Yang uses Zapier to automate roughly 1,200 tasks a monthuschamber.com. These aren’t fantasies; they’re real businesses finding breathing room.
That said, it’s worth asking where to draw the line. Microsoft researchers who analysed 200,000 Copilot conversations found that tasks involving translation, history and writing overlap most with AImarketingprofs.com. They warn that AI supports tasks but doesn’t replace entire jobsbusinessinsider.com. Studies of neurodivergent users point out that over‑reliance on chatbots can erode critical thinkingreuters.com. And the market’s pivot toward safer models shows that organisations are thinking hard about ethics and trust. Shouldn’t small businesses do the same?
A boring prediction
If you’re waiting for AI to become sentient and compose your next novel, you might be disappointed. The changes that will reshape your week are subtle. A Copilot mode that quietly picks a model could influence your productivity more than any flashy headline. A Photoshop tool that erases a bowl of salsa without leaving a green blobtechcrunch.com—and lets you pick which Firefly model to usetechcrunch.com—might not trend on X, but it will make your life easier.
So here’s the question I’m wrestling with: when every tool comes with AI baked in, how will you keep your brand enthusiastic, empathetic, imaginative and clear? Do you chase the latest release because everyone else is, or do you choose the solutions that feel honest and sustainable? Anthropic’s rise hints that trust beats novelty. Microsoft’s smart mode suggests a future where the AI disappears into the background. The real decision isn’t “Which model should I use?” It’s “How do I stay true to my story while all of this technology fades into the wallpaper?” That’s the kind of challenge that keeps me curious.
