May 10, 2026
Midlife Recalibrated
The WeeklyReal AI. Real talk. For women building what’s next.
First, my apologies for getting this out a little late, but I have a pretty good excuse. Mother’s Day. 🥰 I got to celebrate three mothers and be celebrated by my two daughters. It was a special day, and if you celebrated it this weekend too. Happy Mother’s Day. 💐
As for last week: I had a busy one building inside Claude Cowork. I built my Command Center, basically a dashboard to get all my ideas and to-do lists into one organized place instead of scattered everywhere. And I built a Threads Performance Dashboard so I can actually dig into which posts are performing and why, instead of just guessing. There are so many things to build and try right now, and never quite enough hours. 🤪
If you’re new here, welcome. This newsletter goes out every Sunday for women who want to understand what’s happening in AI without drowning in takes from people who are way too excited about it. Once a week, I go through everything that dropped and pull out what actually matters for staying relevant at work or building something on the side. It’s a longer read on purpose. I’d rather give you one good hour than ten forgettable minutes. Grab a coffee.
Short on time? Here’s just the highlights.
- ChatGPT quietly upgraded itself this week. If you’ve opened the app since Monday, you’re already using a new model, one that hallucinates less, gives shorter answers, and now remembers who you are. Worth checking what it actually knows about you.
- PayPal is cutting 20% of its staff because of AI, and new data says this might backfire. A Forrester study found 55% of employers regret AI-driven layoffs. The people hardest to replace weren’t the most senior. They were the ones who could tell when AI was wrong.
- New data confirms that one-person AI-powered businesses are real. Zoom released a report: 30 million solopreneurs in the US, $1.7 trillion in annual revenue, 84% using AI. Top earners use AI at twice the rate of everyone else.
- Threads finally works on desktop. You can now manage your DMs from your laptop instead of your phone, a small change that matters a lot if you’re using Threads to build an audience.
What actually mattered this week
Your ChatGPT just got noticeably smarter, and it now knows who you are
The practical differences are real. GPT-5.5 Instant produces 52.5% fewer false claims on high-stakes topics like medicine, law, and finance, compared to the previous default. Responses are about 30% shorter, which means you get answers instead of essays. And it asks fewer unnecessary clarifying questions before actually doing the thing you asked it to do.
The bigger change is the personalization rollout. GPT-5.5 Instant now pulls from your saved memories, past conversations, and for Plus and Pro users, your connected Gmail to shape responses. When a response is personalized, you can see exactly what context was used and delete or correct anything that’s outdated. That transparency is new and genuinely useful.
For paid users, the previous model (GPT-5.3 Instant) stays available for another three months through settings. After that, GPT-5.5 Instant is what you get.
If ChatGPT has been feeling generic lately, this is worth retesting. The personalization layer is specifically designed to stop it responding like it’s talking to a stranger. Most people have never looked at what’s saved in their ChatGPT memory. This week is a good time to check. (See “Try This Week” below for exactly how to do that.)
PayPal is cutting 20% of its workforce, and the backfire data is worth reading
PayPal is targeting $1.5 billion in cost savings by replacing what those employees were doing with AI and automation. The CEO used the phrase “becoming a technology company again.” Coinbase explicitly said it’s deploying agents to consolidate work. This is the financial sector version of what Cognizant announced in the IT sector last week.
What doesn’t usually run alongside the layoff headline: a Forrester analysis published this week found that 55% of employers regret AI-driven layoffs. The reason that keeps coming up: they cut people who held context, relationships, and judgment that the AI doesn’t have. Separately, researchers from Imperial College London and Microsoft published a warning that AI adoption can actually increase workplace burden, because employees end up managing multiple AI agents and correcting their errors instead of doing the original work.
The honest picture: these layoffs are real, they’re accelerating, and some of them will stick. But the organizations actually seeing results from AI are the ones investing in people who can guide it, not just the ones replacing people with it.
Your job security right now isn’t about whether AI can technically do your job. It’s about whether you’re the person who guides it, corrects it, and makes the judgment calls it can’t make. That’s the distinction companies cutting recklessly are discovering the hard way, and it’s a distinction you can make visible on your resume and LinkedIn profile before someone asks you to justify your role.
New data: AI-powered one-person businesses are real, not a trend
The numbers in the report are worth sitting with. There are more than 30 million solopreneurs in the United States, generating $1.7 trillion in annual revenue (6.8% of total US economic output). 84% of creators report using AI in their workflows. And here’s the specific data point that matters: top-earning solopreneurs use AI tools at twice the rate of lower-earning ones, and achieve 2–5x higher engagement on AI-assisted content.
Of the nearly 3,000 applicants for the Solopreneur 50, 62% were operating active, revenue-generating businesses. The median founding year was 2022. These aren’t people talking about building something. They’re people who built something, with AI, in the last four years, without a team.
The model works. The question is execution.
If you’re building something on the side, or thinking about it, this is useful context to have. When people ask whether the online business thing is realistic, you now have a number to give them: 30 million people are doing it. The ones making it work are the ones using AI. And the gap between “using AI as a shortcut” and “using AI as infrastructure” is exactly where the income difference shows up.
Threads finally works on desktop, which matters more than it sounds
The feature includes your full DM inbox, message search, message requests, and the ability to start new conversations from the Threads website. Group chats are coming in a second rollout, with no confirmed date yet.
One thing worth being straight about: this is a staged rollout, and most people don’t have it yet. Meta hasn’t released numbers on how many accounts are in the initial group, and the only timeline they’ve given is “coming weeks.” If you open Threads on your laptop right now and don’t see a paper plane icon in the left nav, you’re in the queue. You haven’t missed anything.
The reason it’s still worth knowing about: if you’re building an audience on Threads, you’ve been doing all your relationship management from your phone. Every follow-up, every reply to someone who reached out, every connection with a potential client. Once desktop DMs land for you, that friction disappears.
Go to threads.net on your laptop and look for the paper plane icon in the left nav. If it’s there, spend 10 minutes catching up on any DMs you’ve been sitting on. If it’s not there yet, you’re not missing it — Meta said full rollout happens “in the coming weeks” with no hard end date. Worth checking back in a week or two.
The words everyone’s using
Anthropic released a feature this week for developers building Claude-powered AI agents, and the name is doing a lot of work. “Dreaming” is what they’re calling the process where an AI agent reviews its own past sessions, extracts patterns from what worked and what didn’t, and writes them into memory as playbooks for future sessions. The agent effectively learns from its own history without being manually reprogrammed each time. Legal AI company Harvey implemented it and saw task completion rates increase roughly 6x.
You don’t need to build AI agents for this to be relevant to you. What it signals: AI tools are becoming self-improving. The version of ChatGPT or Claude you use today will be noticeably more capable six months from now, not because someone pushed an update, but because it learned from its own work. The tools you’re building fluency with right now aren’t a static target. They’re going to compound.
The U.S. Department of Labor launched a formal contracting process this week to integrate AI skills into Registered Apprenticeship programs nationwide. The selected contractor will build AI curriculum, create training standards, and help employers add AI upskilling to programs that are federally recognized and sometimes publicly funded. This is a federal-level recognition that AI competency is now a workforce basic, being treated the same way computer literacy was 20 years ago.
This won’t change your situation this week. But the federal government putting infrastructure behind AI training is not a small thing. It’s a policy signal that “knows AI” is moving from competitive advantage toward expected. If you’re building AI competency now, you’re ahead of a curve that’s about to get a lot of institutional momentum behind it.
Alongside GPT-5.5 Instant, OpenAI released three new voice models in its API this week: GPT-Realtime-2 (smarter live voice reasoning that can think while it listens), GPT-Realtime-Translate (real-time speech translation across languages, mid-conversation), and GPT-Realtime-Whisper (streaming transcription that captures speech as it’s being said). These are developer-facing right now and aren’t in the ChatGPT app yet.
Voice AI that can translate in real time and reason while listening is about to get embedded into a lot of products you already use: customer service software, meeting transcription tools, communication platforms. If any part of your work involves live conversation or meetings, this is the category to understand before it shows up in your company’s next software update.
Connect the dots to your actual life
If you’re in corporate or thinking about your next move…
The PayPal story paired with the backfire data is the one to sit with this week. Here’s the frame that’s actually useful: the people proving hardest to replace after AI-driven layoffs aren’t the highest-paid or most senior. They’re the ones who could tell when an AI output was wrong, knew what to do when it failed, and understood the human context a model can’t read. That’s a describable skillset, and right now, most people aren’t describing it.
This week: update your LinkedIn with one specific, concrete example of how you guide, evaluate, or correct AI in your role. Not “familiar with AI tools.” Something like: “Review AI-generated analysis against source data before stakeholder delivery” or “Evaluate AI-drafted communications for tone and accuracy.” That distinction, between using AI and being able to assess it, is exactly what the companies regretting their layoffs wish they’d kept.
If you’re building something on the side…
The Zoom Solopreneur data is your permission slip, but it’s also a prompt. The report found that top earners use AI at twice the rate of everyone else, and that they’re achieving 2–5x higher engagement. The difference isn’t the tools. It’s how they use them.
This week: start treating your AI usage as data. Every time AI saves you an hour, note it. Build a short list of what you actually use AI for in your business. That list becomes content (your audience wants to know this), it becomes your product positioning, and it becomes the credibility signal that makes people trust you as someone who has actually figured this out, not just someone who bought a course about it.
Find out what ChatGPT thinks it knows about you
Open ChatGPT on the web. Paste this prompt, fill in the bracket, and hit send. Review what comes back: correct anything wrong, add anything missing, and delete anything irrelevant. The new personalization layer uses this to shape every future response.
Already prompting well? Use the new memory transparency to understand why.
GPT-5.5 Instant added a small info icon on responses that were personalized using your memory or past chats. Click it and you can see exactly which saved context the model pulled to generate that specific answer. If you’ve been noticing that some prompts land better than others and wondering why, this is the diagnostic to start using. Trace what context was used, refine what’s in your memory settings, and you’ll start getting consistently better responses without needing to rebuild your prompt every time.
The PayPal layoffs and the Zoom Solopreneur report dropped within 48 hours of each other. Nobody’s reading them together.
One company is eliminating 4,760 people because AI can supposedly handle the work. Thirty million people are already building their own things with the same tools, doing it well enough to generate $1.7 trillion in revenue. Some of them started because they saw a layoff coming. Some started because they got one. The common thread isn’t a certain background or a certain skill level. It’s that they stopped waiting to see what would happen to them and started deciding what they were going to build with the same technology everyone is so afraid of.
Until next Sunday,
Jess ☕
Midlife Recalibrated