2025 in review

5 AI features worth paying for

Hello You & AI community—

I'm re-launching the You & AI newsletter. I won't try to keep up with every AI announcement because no one needs that. Instead, I'll share what's changing in my own use, what I'm learning from teaching & working with clients, and thoughts on the implications of ~all this~ for education & the future of work. You'll also hear about new offerings, and I'll keep sharing my travel schedule because it's been fun meeting you & your connections in person.

I want to kick off the newsletter with a reflection on the AI advances in 2025 that most changed my work. So not things that already existed like project instructions & persistent memory or things that already existed but were new to me like coding in html, JavaScript, and Python. But truly new releases in 2025.

After I made my list, I asked AI for the 5 biggest AI headlines of 2025, and there was zero overlap. Spectacle dominated the headlines: real-time voice mode that feels like Her, video generation that stunned social media, and endless demos of "agents." The things that actually changed my work were less splashy.

The five AI advances that mattered most for my work in 2025:

1. o3 (April 2025). OpenAI's o3 was the first reasoning model that felt genuinely smart to me—higher quality answers, more capable of reaching for the right tools like Python or web search at the right time, and better at executing multistep processes in one go. It made survey analysis reliable, returned real links instead of hallucinated ones, and even helped me fix my sink–diagnosing the issue, finding the exact replacement part I needed, and generating a shortlist of plumbers with a service request to send to them. o3 has since been superseded by better models from every major lab, but everything else on this list is downstream of better models.

2. Claude in Excel (October 2025). After 10+ years in Sheets, I switched back to Excel entirely because of how much value I get from Claude in Excel. I started my career as a Bain analyst, and Claude in Excel makes it feel like I have a Bain analyst working for me. Only it's faster, even more capable, and costs less than $30/month. I built the best financial model of my life with this tool, cleaned up messy You & AI participant data with it, and used it to quickly categorize a slew of expenses. I'll likely run a deep dive workshop on Claude in Excel soon, and if you're interested, you can indicate that here.

3. Claude Code (~August 2025). Claude Code can write scripts, manage files, and build tools you'd otherwise need a developer for. For example, I had it build a script that runs every night at 8pm, renaming that day's screenshots so I can actually find them later. If I had started using it earlier in the year, it would have been #2 on this list because it's much more powerful than chat-based AI: it can take actions, not just recommend them; it can automate actions & workflows; and it can scale these actions to datasets & workflows of great size. But there's a learning curve to it, and in 2025 I was still on the learning curve, so it was my #3. It's in my plan for the year to develop & teach a course on Claude Code.

4. Claude Skills (October 2025). Skills let you teach Claude a task once, then it applies those instructions when you ask it to or when it perceives it would be useful. For example, I write this newsletter in plain text, and Claude converts it to Mailchimp-ready HTML with my branding using a Skill that I built. Same with certificates for workshop participants and marketing collateral–both Skills. Anthropic has already announced that it will be launching enterprise toggles that will make it easier to share Skills with others on your team.

5. Connectors (April 2025). Context makes AI smarter, and connectors enable AI to look for context to help fill in gaps. Connectors link AI to your existing information—documents, email, payroll, etc.—so relevant context can be pulled into your chats. This can either be on your request (e.g., "What do we say about customer acquisition cost in the latest pitch deck?) or because the AI senses a gap in its knowledge and proactively seeks context to fill it (e.g., You ask for a one-pager on your organization but don't give context in the prompt or attach any files). To share a connector use case from a recent You & AI participant, Lateefah Durant of CityWorks reports that "[For] our team reflection meeting, I used connectors to analyze what I focused on over the past four weeks. The tool answered the prompt in five minutes and successfully identified how my work this month linked to my quarterly priorities, which I had stored in memory."

Those are my top 5. What would you add or change?

Cheers,

Sarah

(All em dashes are my own.)