The Last AI Update of 2025
Agents, accountability, and what HR needs to figure out before January
Welcome to FullStack HR, and an extra welcome to the 48 people who have signed up since last week.If you haven’t yet subscribed, join the 10100+ smart, curious leaders by subscribing here:
Happy Monday!
I’ve been really consistent with these updates. Then December crept up on me. You know how it is…
But before the holidays kick in, here’s the final update of the year. The regular newsletter will still go out twice more, before Christmas and between the holidays. This update takes a break until January 9.
So this is the last one for 2025. Let’s go.
PS. Reminder about this one, kick-start your year with a fantastic course - it’s my full AI adoption playbook!
The course starts on the 8th of January.
🧠 Topics I’m engaging with
How AI Is Transforming Work at Anthropic
Anthropic shares what usage actually looks like inside an AI lab. Employees self-report using Claude in about 60% of their work and claim roughly a 50% productivity boost, more than double what they reported a year earlier. Key insight from the article is what I’ve been talking about for the last year as well: the big skill isn’t prompting. It’s delegation + verification. Claude enables people to become more full‑stack, for example backend engineers building UIs and non‑technical staff handling debugging and data analysis they previously avoided.
I really like that we get more data points like this all the time. It’s yet another useful “real adoption” datapoints. And even though I don’t think that you could copy-paste the approach from Anthropic, you could let it inspire you. And for me, the takeaway isn’t “you’ll get 50% gains.” It’s that gains require role clarity, guardrails, and follow-up loops. (Hey, sounds like HR could play a part in this transformation…)
We Let AI Run Our Office Vending Machine. It Lost Hundreds of Dollars.
WSJ ran a simple real-world agent test, a follow up to Project Vend that Anthropic ran a while back. Give an AI autonomy in a narrow business context of running a vending machine, ordering inventory, setting prices, and interacting with humans. Just as with project Vend, it goes wrong…fast.
The obvious reaction is “agents aren’t ready.” The smarter reaction is “great data point towards something useful”. It’s supereasy to laugh at these projects and call them failures. But are they really? To some extent, yes, as they did not work exactly as thought, but they serve as data points back to Anthropic, and as such, they are great.
HR’s Existential Moment: Reinvent or Fade Away
Great piece from Brian Elliott and a clear take on why people teams matter more when work changes fast. According to Brian (and I support this view) AI can either make HR irrelevant or elevate us into the strategic engine of organizational transformation, depending on whether people leaders reinvent their work around business impact, product thinking, and skills-based strategy.
No surprise that I agree with the direction. If AI makes output cheaper, what becomes scarce? Usually judgment, trust, coordination. That’s HR territory. The teams that own that question will stay relevant. The ones waiting for someone else to fi
gure it out won’t.
Designing Your Workforce for Agents
A practical HR framing for agentic AI by : where to start, what to redesign, how to avoid breaking trust. The recommendation: push for narrow, low-risk domains first and insist on visible governance.
Otherwise, you get shadow agents. And no organizational learning. Workforce planning now has to include “agent capacity” and “human oversight capacity.” Job design shifts toward orchestration and exception handling. And the training need is role-specific, not generic “AI literacy.” Start small. Learn fast. Scale what works.
⚙️ Tools to try
Sanas's “bring me up to speed before this meeting” is bonkers.*
*Sponsored by Sana
🔁 Important updates
Official launch post for GPT-5.2 that launed two weeks ago (told you I was late…)
Gemini API changelog: Gemini 3 Flash Preview
Gemini has rolled out TWO models since I last posted (Gemini 3 Pro and Flash) both are GREAT. Test them during the holidays!
HR Conferences 2026
Hung Lee has once again crowd-sourced a list of conferences to attend in 2026, and I’ve put my best vibe-coding skills to use to make a better front-end to the Google sheet.