.png)
October 4, 2026
72% of AI users have heard of AI agents, but only 41% have tried one. If you are standing in that gap, this newsletter is your map.
In our FlipWork™ Sprints, we work with leaders and corporate professionals who are eager and driven to become more agentic human leaders. Lately, with all the new bots from ChatGPT, X, and Meta, one topic keeps coming up.
What should we use? Should we try all of them? Is it worth the expense?
Some cohort members felt confused and overwhelmed. Then came a few aha moments, when we shared how the tools can split the work. Muse for personal life. Grok Bot for work. Claude for separating out different kinds of tasks, like writing.
Here is how I do it. I use all of them and play to their strengths. Claude is where I do most of my writing and learning design. I tried ChatGPT for writing and didn't like it, so I'm staying with Claude. I use ChatGPT for the scientific logic and math behind our FlipFactor™ diagnostic. FlipWork™ runs on Google Workspace, so I use Gemini when I need current sources, since it draws on Google Search, and sometimes to compare or polish a final draft. Gemini 4 Argon holds a preliminary #1 on Arena's human-voted creative writing board, with Claude a close second. Still, Claude is easier and saves me time, so I reach for it more. Grok Bot handles admin and operational work. Muse is for my personal life, and I'm about to test it on math and data too. On Vals AI's Finance Agent benchmark, Meta's Muse Spark models ranked #3 and #4, ahead of every Claude and GPT model, at a fraction of what top Claude models cost per test. That's a finance and data analysis test, not pure math, which is why your own testing matters more than any ranking.
Use cases like these showed our cohort there is nothing to fear. Ironically, most of them were already using agents. If you have ever set up a scheduled task, or automated or semi-automated something for yourself, you were, in essence, using an agent. What's new is that agents are now out-of-the-box and plug-and-play.
If you are asking the same questions, I want you to hear this first. You are not behind, and you have nothing to feel guilty about.
This feels overwhelming because it is overwhelming. You know that Friday night feeling when you finally sit down to watch a movie? Netflix, Prime, Hulu, Paramount, Max, Disney. You spend more time scrolling trailers than watching a show.
My Dad knows that feeling better than anyone. He loves westerns. Real westerns. And he gets so frustrated when he can't find the shows he wants to watch on TV. All those options, and hardly any good westerns. The new-age, modern ones? Too modernized, too cheesy. And the sci-fi westerns? Definitely not. (That last one still makes me laugh.)
All that scrolling for so little reminds me of a saying he used while shelling pecans and walnuts.
"Mucho ruido y pocas nueces." (Lots of noise, few nuts.)
Cracking them makes a lot of noise, but you don't get much nut meat unless you crack it just right. His goal was always to get the whole half out. Sometimes it was a lot of work for very little, bits and pieces instead of the whole nut. He loved eating pecans, and that's when he'd say it: a lot of noise for very little nut. Same with sunflower seeds, which he also loves. All that cracking and peeling for a tiny morsel.
That is what AI agents feel like right now. In about seven weeks, three of the biggest names in tech each launched a personal agent. The launches, the demos, and the headlines are the noise. Crack an agent the wrong way and you get bits and pieces. Crack it right, and you get the whole half: the few things each tool can actually do for you, in your life and your work.
In a past newsletter, Your Favorite AI Is Not Your Best AI, I said every tool has its own zone of genius. Agents are no different. IBM describes an agent as a system that autonomously performs tasks by designing workflows with available tools. In plain terms, a chatbot answers. An agent does.
1. Most people have heard the word. Far fewer have tried one. Menlo Ventures and Morning Consult surveyed 5,067 U.S. adults for 2026: The State of Consumer AI, published September 16. Among AI users, 72% are aware of AI agents, 41% have tried one, and 24% use one regularly. Getting started is the gap.
2. Many people are already using agents without calling them that. The same survey found that 32% of AI users have let AI act on their behalf at least once without a final sign-off, like completing a purchase or sending an email. It is self-reported, so hold the number loosely.
3. The personal agent wave arrived all at once. SpaceXAI launched Grok Bot on August 11. Meta introduced Muse on September 8. OpenAI introduced dots on September 29. On October 1, The AI Daily Brief reported that Muse reached 3 million weekly active users in its first three weeks.
Put those together and you get a lot of people who know the word, fewer who have tried an agent, and plenty already using one without the label, all while three new agents landed in seven weeks. No wonder it feels like noise.
Prices are as of October 4, 2026.
Muse (Meta)
Grok Bot (SpaceXAI)
Dots (OpenAI)
ChatGPT scheduled tasks and Work (OpenAI)
Claude (Anthropic)
Gemini (Google)
Notice the pattern our Sprint cohort spotted. Muse leans personal. Most of the rest lean work.
Menlo also found that people who pay for AI are five times as likely to use agents, 65% versus 13%. Paid access shapes who gets to practice.
Why we believe it: The new bots arrived with names, avatars, and launch events. Nobody threw a launch party for your ChatGPT reminder.
Reframe: An agent is a job, not a brand name. If it works toward a goal without you typing every step, it is acting as an agent. IBM's simplest example is a thermostat that turns on the heat at 8 PM every night. A ChatGPT task that checks the news every morning and sends you a summary does more than that, so like our Sprint cohort, many of you have already crossed the line.
What to do: Open the Scheduled page in ChatGPT or Claude. If you have tasks there, you already run an agent. If not, you are one sentence away.
You don't need all of them. As I've shared before in my newsletters, the rule is: start with two, not five. One for personal, one for work.
I'm practicing what I preach. This week, I'm testing Grok Bot to help me write this very newsletter, to get a feel for how well it does. I've already given it my personal Markdown files with context on how I write and what I prefer, so it knows me better. Then I'll grade it against Claude and ChatGPT, and I'll test Muse next on the numbers, given its price. I'll share how they grade out.
Run this 30-minute test:
Prompt to steal:
I am new to AI agents. Here is one task I repeat every week: [TASK]. First, explain in plain language the steps you would take, what access you need, and where you would stop and ask me. Then do only the first step. Do not send, buy, delete, or post anything without my approval. If this should run on a schedule, suggest one: [HOW OFTEN, E.G., EVERY MONDAY MORNING].
1. Set up one scheduled task. It works on ChatGPT's free plan. Try "Every Friday afternoon, give me three things I should prepare for next week based on [YOUR PRIORITIES]."
2. Make two lists. Personal errands and work tasks you would love to hand off. Match each list to a tool.
3. Ask before you connect. Find out which agents your company has approved.
Start every agent with draft permission, then let it earn send permission. Grok Bot and Dots let you set rules for what needs your OK. For two weeks, let the agent read and draft while you click send. Trust is built in small steps, with people and with bots.
Remember the three questions from our Sprint? What should you use? Start with two, one personal and one work. Should you try all of them? No. Is it worth the expense? Start with the free and low-cost doors, and pay only when a tool earns its place.
Go back to that Friday night on the couch. The fix was never watching more trailers. It was picking one show and pressing play.
Agents work the same way. You need one personal errand, one work task, and thirty minutes.
My Dad is still hunting for a good western. Bless him, he is not settling for the cheesy ones, and you don't have to settle for bits and pieces either. Mucho ruido y pocas nueces. Take your time, crack it right, and get the whole half out.
One more thing. Grok Bot helped me write this newsletter. Could you tell? Hit reply or email me, and help me decide whether I keep using it for the newsletter or stick with what I know.
♻️ Share this with someone who keeps asking which AI agent to try and hasn't pressed play on one yet.
¡Hasta la próxima, un abrazo fuerte! (Until next week, a big hug!)
