I’ve been noticing a pattern lately with how we talk about new AI releases.
Every Tuesday, it feels like a new "world-changing" model drops. My LinkedIn feed fills up with "It’s over for [insert industry]" and "10 prompts to change your life."
But then I sit down at my desk on Monday morning, and I realize most of that noise doesn't actually help me get through my to-do list.
I’ve spent the last few months filtering through the constant updates to figure out what is actually helping agencies run better, and what is just a shiny distraction. In my experience, the tools that matter aren't always the ones with the loudest marketing—they're the ones that quietly solve a friction point in a boring process.
The Shift From Chatbots to "Large Action Models"
For a long time, AI was just a very smart box we talked to. We spent most of our energy learning how to "prompt" it to get a decent paragraph of text back.
The most significant shift I’m seeing right now—specifically with things like Anthropic’s "Computer Use" and OpenAI’s "Operator" concepts—is the move toward the AI actually doing the work.
Anthropic recently released a capability where Claude can essentially "see" a computer screen, move the cursor, and click buttons. While it’s still in beta and a bit clunky, the implications for agency operations are massive.
Instead of just asking an AI to "write a report," we’re moving toward a world where you tell the tool: "Go into our CRM, pull the last month of lead data, cross-reference it with our ad spend in Meta, and drop a summary into the Slack channel."
It's moving from content generation to task execution. That is a distinction worth paying attention to.
Project Management That Actually Thinks
I think we’ve all felt the pain of "Project Management Overhead." You spend so much time moving cards around in Trello or updating statuses in Notion that you barely have time to do the actual work.
I’ve been watching how tools like Height and Linear are integrating AI, and it’s one of the few areas where the hype meets reality.
Height recently launched "Copilot," which doesn't just suggest tasks; it handles the "janitorial" work of management. It can automatically deduplicate tasks, categorize bug reports, and even draft responses to team members based on the context of a project.
For a small agency, this is huge. Most of us can’t afford a full-time project manager for every account. If an AI can handle the sorting, tagging, and status-updating, it frees up the team to actually be creative.
It’s not revolutionary in a "the world is changing" sense, but it is revolutionary in a "I get an extra hour of my life back every day" sense.
High-Fidelity Research with NotebookLM
If you haven't played with Google’s NotebookLM yet, I’d suggest giving it a look.
Most LLMs struggle with "hallucinations"—making stuff up because they are trying to be helpful. NotebookLM fixes this by being "grounded" in the documents you provide. You upload your PDFs, your client's brand guidelines, or your meeting transcripts, and it only answers based on those sources.
The "Deep Dive" audio feature, which creates a podcast-style conversation between two AI hosts about your data, got a lot of social media attention. It’s cool, but it’s a bit of a gimmick.
The real value is the "Source Grounding."
When I’m onboarding a new client with 50 pages of messy documentation, I can drop it into a notebook and ask, "What are the three biggest pain points mentioned in these transcripts?" and get an answer with citations.
It turns a three-hour reading task into a fifteen-minute synthesis task. That is a real-world win.
Video That Doesn't Look Like a Nightmare
We’ve all seen the early AI video—the weird six-fingered hands and the melting faces. It was impressive for five seconds, then it was just creepy.
However, the recent updates from Kling, Luma Dream Machine, and Runway (specifically Gen-3 Alpha) have crossed a threshold. We are seeing tools that can maintain "character consistency."
For agencies doing social media content or mood boarding for clients, this is a massive shift. You can now generate high-quality B-roll that actually matches your brand aesthetic without a $10k production budget for a 15-second clip.
Is it going to replace a professional film crew for a national TV spot? No.
But for the "middle-of-the-funnel" content that every brand desperately needs—the stuff that usually gets skipped because it’s too expensive to shoot—it’s a game-changer. (I promised myself I wouldn't use that word, but in the context of production costs, it’s hard not to).
Making Content Actually Sound Like You
The biggest complaint I hear about AI-generated content is that it sounds "AI-ish." It’s too bubbly, it uses too many emojis, and it sounds like a generic marketing brochure from 2005.
The new Custom Instructions and System Prompts features in ChatGPT and Claude (specifically "Projects" in Claude) have finally started to solve this.
I keep coming back to the idea that an AI is only as good as the context you give it. By using Claude Projects, I can upload my previous blog posts, my brand voice guide, and my "list of words I hate."
Now, when I ask it to help me outline a post, it doesn't start with "In today's fast-paced digital world." It starts where I would start.
The tool itself hasn't changed as much as our ability to "tether" it to our own identity. If you're still using a blank chat box for every new task, you're doing it the hard way.
What to Ignore (For Now)
It’s just as important to know what to ignore.
The "AI Search" wars—SearchGPT vs. Perplexity vs. Google—are fascinating to watch, but they shouldn't necessarily change your agency strategy yet. Yes, SEO is changing, but the fundamental need for high-quality, authoritative content hasn't. Don't throw your SEO strategy out the window because of a new search interface.
I’d also be wary of any tool that promises "Fully Automated Sales."
I’ve seen a lot of "AI SDR" tools lately that promise to send thousands of personalized emails. The problem is, everyone else has those tools too. Inboxes are being flooded with "personalized" AI noise.
In my experience, as AI makes "quantity" cheaper, "quality" and human connection become significantly more valuable. Use the AI to research the person, but write the email yourself.
A Simpler Way to Think About This
Whenever a new tool drops, I try to ask myself one question:
"Does this reduce the number of steps between an idea and its execution?"
If a tool adds a new platform I have to manage, a new subscription I have to track, and a steep learning curve for my team, I usually pass.
But if it replaces a manual spreadsheet, or if it helps a junior designer get a head start on a mood board, it stays.
The bit most people miss is that AI isn't a replacement for your expertise. It’s a high-speed bicycle for your brain. It gets you where you're going faster, but you still have to know how to steer.
I’m curious—which of these have you actually tried in your daily workflow? Are you finding the "computer use" stuff genuinely helpful, or is it still a bit too experimental for your liking?
I’ve found that the best way to learn this stuff isn't by reading the manuals, but by trying to solve one very small, very boring problem with it. Start there.
I write about AI tools, business, and unconventional working life at steventann.com. Come say hello.
About the Author
Steven Tann is an AI consultant, author of "You're Selling AI Wrong", and founder of SalesM8. He writes about AI, sales, and running a business from a narrowboat on the English canals. Connect with him at steventann.com.
