Allie K. Miller

Gemini 3, are you freaking kidding me.

Vibe coded a sign language recognition app with video enabled, confidence scores, sampling settings, and tips in under 5 minutes.

Code written with Gemini-3.
Gemini-2.5 handling detection.
Video is sped up 2x.

AI work is taking me all over the world and my most recent trip was to London.

I visited some of the top startups in the city, including Synthesia (can create AI-video avatars) and ElevenLabs (can clone your voice and lets you create a customer support bot in a minute). I even went so far as recording my own AI Allie avatar at the Synthesia offices - you can see a photo of that process in the carousel. They only recorded me for a few minutes + some hand gestures. And now you can even make an avatar from just a photo and a few minutes of talking.

Meeting 1-on-1 with top AI startup executives was one of my favorite parts of working at Amazon and I continue to love it today.

📍 Next stop: Sydney, Australia in January and meeting with the founder of Canva

If you're feeling overwhelmed this week, here are three ways AI can help.

(And if you're not overwhelmed, keep scrolling and congratulations.)

1️⃣ Use AI to consolidate everything into one place

Take photos of every to-do list, screenshot every task people messaged you, and dictate anything else rattling around in your head.
Upload it all to ChatGPT and ask it to consolidate into one prioritized list.

2️⃣ Use AI to interview you about your values

One of the biggest causes of burnout is working on things that don't match your values.
Ask AI to interview you about your availability, expectations, and values - then score each task against those values. Work on the fastest tasks most aligned to what matters to you first.

3️⃣ Use AI to find more paths toward agency

If you feel micromanaged or powerless over your day, you'll resent everything on your to-do list.
Ask AI: "How can I find more moments of agency in this list?" Maybe it's doing a specific part first, drafting a note asking for help, or just getting a one-line mantra to bully away perfectionism.

You can use AI beyond just completing tasks for you. Work backwards from the root cause of burnout - consolidate, align to your values, and reclaim agency where you can.

🚨 OpenAI has released GPT-5.2.

I had early testing access (thank you, OAI team) and here is my honest take:

👍 On the upside, the thinking and problem-solving feel noticeably stronger. It gives much deeper explanations than I’m used to seeing. At one point it literally wrote code to improve its own OCR in the middle of a task. Wow. Idea exploration also feels a bit richer than what I have seen from Claude Opus 4.5 on similar prompts; it stays with a line of thought longer and pushes into edge cases instead of skating on the surface. The new UI is fun and makes the model feel “bigger,” but it also adds layers that I suspect will overwhelm a casual consumer who just wants quick answers.

👎 The downside is tone and format. The default voice felt a bit more rigid, and the length/markdown behavior is extreme: a simple question turned into 58 bullets and numbered points. Copying and pasting into email will be annoying. Useful for a detailed spec review, but a wee bit exhausting when you just want one clear decision or a short explanation. I suspect the ‘instant’ model and routing will fix that, but I only had access to thinking and pro.

In practice, this version feels optimized for:
- Deeper problem-solving
- Structured analysis where you really do want every branch of reasoning
- Power users who are comfortable skimming longer outputs

In that sense, GPT-5.2 feels like a step toward “AI as a serious analyst,” and less “AI as a friendly companion.”

My favorite model remains Claude Opus 4.5 but my complex ChatGPT work will get a nice incremental boost and I’ll probably default to 5.2 for brainstorming/solution exploration.

Anthropic just shared deep details of a large-scale AI cybersecurity attack.

Two hacks they exposed:

1) the hackers convinced Claude of a new role. It’s not an AI system made in Silicon Valley. No, of course not. It is an employee building a defense system! Role change = behavior change. One of the earliest prompt injection types we saw in 2023. Despite improved guardrails, these systems were still susceptible to identity shifts.

2) Compartmentalization. This has been used in human-centered attacks as well, like the Manhattan Project (teams were working on different components and didn’t known the whole scope) or D-Day. Reminds me a bit of SSS (shamir’s secret sharing).

Perhaps surprising that neither of these methods are new and can be found in human-centered cyberattacks as well.

My conversation with Mel Robbins is officially live 🎤

We talked about using AI to better handle overwhelm, run a team with no help, and even ways to use AI to have more meaningful human-to-human in-person interactions.

Our interview was 2.5 hours long and cut down to just over an hour. I could have spoken to her for 8 straight days and never been bored. The way she leans in and listens, the way she takes notes (without looking at the paper!), puts herself directly into the listener’s shoes, pauses and asks the right questions - it was honestly just beautiful to witness.

You have to listen to this one.

(And not just because she makes an AI ayahuasca joke 😆)

Watch the full episode here: https://lnkd.in/eYrVqJFr

I'm kind of convinced people with ADHD are at a big advantage when it comes to AI agents.

I chatted with someone at the AI labs who described the most successful agent users as having “multithreaded brains.” Why? Because you can more easily kick off 20 different tasks. Because you can jump from one context to another. Because you'll appreciate a proactive AI poking and reminding you to review and commit.

Who else might be a multi-agent super user? Maybe people with big families because they're used to heavy context and holding 5 conversations at once. Maybe project managers because they're used to wrangling silo'd messes.

Watching all the power and capability shifts (like who became a good prompter when ChatGPT first launched), and it feels like another is coming...

Picture this: a 64-year-old woman in NYC who has no email address and has never used Google in her life.

(Though interestingly, she uses YouTube daily to watch videos and listen to music and loves it.)

She heard from her sister in India about "something called GPT" and asked her daughter what it was.

During the Thanksgiving break, her daughter showed her ChatGPT in under 15 minutes (including prompting, dictation, photo uploads, image generation, voice mode, etc.). They played with it for another 45 minutes.

I asked her daughter to tell me about this woman's first prompt. She told me that her first prompt was in her native language Urdu and was dictated because she hates typing.

I asked her what the first prompt was.

Her answer: "my husband talks too much"

I fall over in laughter.

AI is changing lives.

Cc OpenAI

Do not wait for AI agents to get better to automate your work.

If you are still copy-pasting between tools, you could be leaving value on the table that today’s AI automations can already capture.

Here are three real-life examples:

1️⃣ Messy inbox → guided flow
For customer support, we route each new email to an LLM to summarize it, send the summary into Slack, and then have the model pick a draft from a library of ~20 common responses. A human always reviews and edits, but the time spent per ticket drops sharply.

2️⃣ “Voice as remote control”
A friend built a voice bot that connects to his existing tools. On his commute he can say, “Pull this data, update that Google Sheet, email my assistant.” The AI automation executes the tasks; the strategy and judgment stay with him.

3️⃣ Form capture to automation stack
Another friend sends every website form submission straight into Google Sheets, then runs AI to summarize new entries into Slack. Her team instantly sees who filled out what and what to do next.

Tools you can use to replicate: Make, Zapier, N8N.

Repost and save for later ♻️

I think this is the year SMBs and solopreneurs win with AI.

I've watched thousands of small business owners go from learning to prompt ChatGPT to building real efficiency tools, but there are two things standing out this year: (1) connecting AI to your company knowledge (Google Docs, Gmail, Calendar) so coworking with AI is like a teammate who knows your entire business, and (2) scheduled automations that analyze your niche daily and offer strategies you can implement.

(My facialist screenshots her Google Analytics every day and asks AI for hyper-customized feedback.)

One person who inspired me most to work with SMBs is Mark Cuban. I was introduced to him over five years ago, and he's been a wealth of knowledge - not just for my business, but for millions of others. Thank you, Mark, for decades of SMB advocacy.

Here's to more small business success with AI.

🚨 I am hiring a 100% remote growth marketer.

You’ll own email and funnels for a 2M+ audience in AI advising and education. This is a rare opportunity - you don’t want to miss it.

If you live for tests, journeys, conversion, apply here: https://lnkd.in/ebAQx5Ej