Imagine your Monday morning in the very near future.
You run a small business. You have two calls before lunch, a half dozen projects that need decisions made, and you’ve got to squeeze in time to review that financial report before a crucial client call in the afternoon.
Luckily, you’re totally prepared. As you drove to the office this morning, your assistant Jolly updated you on what changed since yesterday, what you need to keep in mind for today, and what work is falling behind. You and Jolly talk through some info that needs to be added to a couple of projects, and you even managed to respond to all of your emails before you finished your drive. Jolly even reviewed that spreadsheet you had put off taking a look at, and has produced a quick report for you to read that summarizes the most important changes that you should take a look at.
You get to work and you’re pretty much caught up. You grab a coffee, sit down at your desk, and turn your attention to the creative and strategic work you love. The tedious and repetitive managerial aspects of work that you always tolerated (but never loved) have become a lot more manageable, which has freed you up to do whatever you think makes sense.
This fictional morning is rapidly becoming our new reality.
September was the month of agents with always-on computers. Earlier this year, I wrote The Personal Computer That Never Sleeps, an article about keeping separate machines available for agents to work while you’re working and living your life. My example from this summer involved always-on Macs and Windows machines, but it became very clear that this was the direction things were headed with agentics. Now that AI can operate a computer, the question is where it should live and who should control it.
The Pattern
All of the latest agentic tools that are hitting the mainstream have a few things in common, and all of them are game changers for how we naturally work. How we work has been changing a lot this year, and a confluence of different technologies and concepts are all merging into a new product class that is going to be both extremely popular and useful.
This pattern really kicked off with OpenClaw and Hermes earlier this year, and before that Zo Computer, but September was officially the month that always-on computer-use agents became mainstream.
This new working and living paradigm is easy to understand. Your assistant has a real place where it can work. It can use applications you already use. You can close your laptop and go on a walk, and it can keep going. You can iterate using natural language, and have a persistent conversation that moves multiple workstreams forward in parallel.
For entrepreneurs and business owners, three things make this combination important:
Context is part of the assistance: The more your assistant has to work with, the more holistic understanding it can have over what is going on with your work and life.
Work can continue anywhere: your assistant has a persistent workspace for its files and ongoing tasks. Scheduling and automation are limited only by your imagination.
Delegation happens through natural language: you can explain what you want or how you think something should look, and iterate on it just like you would with a person. You can change direction without manually navigating every intermediate step.
Work has shifted from typing and clicking to voice dictation and delegation. Your relationship with your computer, as a result, can and will change.
Grok Bot
Grok is a clear example of this shift.
You can create a bot with a particular responsibility, give it context and what it needs, and communicate with it purely through messages, voice, or even a call. It works across the tools you give it access to, retains useful context, and can be scheduled to do whatever you want it to do. It has a persistent Linux instance, and can install tools to do whatever you need done.
Grok can work alongside other bots, sharing context and handling independent tasks in parallel. This isn’t too dissimilar from how parallel agents work on platforms like Codex or Claude, but the accessibility and form factor (an app on your phone you can talk to) is decidedly different.
Relating info across tools, and then using that information to generate or send new info is exactly what people do at organizations.
Your phone becomes a portal to direct the work, rather than acting as a machine that performs the work. The cost barrier to having a seriously good virtual assistant dropped this year, and the technical barrier has now nearly evaporated.
You don’t need a custom DigitalOcean droplet with a custom build of Hermes via Telegram to have an always-on agent that can assist you. It is okay to not know what that is! Not everyone needs to personally maintain an agent server.
Grok is clearly geared towards professional use. It is straightforward: an always-on agent that has fantastic tool-usage and stays contextually aware.
Meta’s latest product, Muse, has taken a different approach.
Muse
The release of Muse surprised me. Meta has taken a much more personal approach with their new agentic product. It runs in a dedicated cloud environment called Muse Secure VM, which has its own browser. Muse can take on tasks, work on longer-term goals, remember your preferences, and keep working when you close the app. You decide what level of access it has, and it asks for permission for performing certain sensitive actions.
Grok’s product positioning is ‘bots with jobs’. Muse is clearly being marketed as a personal assistant that fits into your entire life.
Muse can easily be used for work, and Grok can just as easily be used for your personal life. The significant difference is the interface. Muse has an info feed, relevant to your life, can track goals, and has a space for storing ideas. Muse can link to your Meta accounts you already use, making autonomous activity across WhatsApp and Instagram a new reality.
Voice Connects Everything
I keep coming back to this. The usage of voice is such a practical interface for computing. Every major platform this year has shipped with audio transcription and conversational interfaces. The ability to be able to go take a walk and get meaningful work done that would normally be done at a computer is finally a reality.
OpenAI’s September 23rd update is particularly meaningful here. GPT-Live now supports plugins and connected apps. This was a major gripe I had earlier this summer, and I’m thrilled that the platform finally supports it. Voice in Work can initiate tasks with documents, connected applications, and even browser use. OpenAI’s always-on agent platform is rumored to be released soon.
What meaningfully changed this year is pretty clear: AI’s ability to reason across multiple pieces of data or sources is drastically better. You can now say ‘Take a look at this spreadsheet, check on the status of this project, and see what conversations happened in Slack about it’ and you’ll get back a coherent response.
We all have collective information overload, from too many emails at work to social media and news. If intelligently used, this tech has the potential to reduce our digital consumption and focus on what really matters.
I’m not saying that screens are going to go away. I think that this has the potential to free up your time so that you have the energy to do your best work: the creative and genuinely unique work that only people can do. Tediously moving statuses in Asana or neurotically checking Slack doesn’t have to be the modus operandi.
Your Digital Chief of Staff
I’ve worked as a Chief of Staff. You’re expected to be on top of what’s happening, maintaining context, prepping for decisions, and ensuring the team has alignment.
This year, AI has become way better at this, and that will increasingly become the case. But your taste and judgement are arguably more important than ever. AI is an amplifier and a rocketship for your ideas. It can just as easily guide you in the right direction as it can lead you down an unintuitive, messy path. The time you save from moving boxes around on your computer can now be spent refining your taste and discernment.
Cognitive offloading can lead to workslop. Workslop can and will erode trust with the people you work with. This is why maintaining a high standard of excellence is so crucial, especially when offloading tasks to AI. Start slow, and mold the agentic system you use into something that works the way you think.
More autonomy for software is only valuable when it creates more agency for the person using it.
Learning how to effectively delegate using AI is a skill that can be grown and nurtured. Just like typing or having a strong intuition for phishing attempts, it is something that takes time to foster.
Work is changing, and the tools we use for working are changing quickly. I’m excited to see how these new products develop and what adoption ends up looking like over the next few months.
These are exactly the kinds of problems I’m working through now with my consultancy: helping businesses figure out where AI can genuinely make work easier, what should stay firmly in human hands, and how to connect the tools they already use without creating an even bigger mess.
If you’re running a business and trying to figure out what any of this should actually look like in practice, I’d love to talk.