Rethinking productivity with AI & Alice
Reflecting on a year of significant achievements in our quest for the ultimate productivity workflow, harnessing the power of automation and AI.
You probably don't know this yet, but I'm obsessed with productivity. I basically have to be, as I run several companies at once, employing dozens of people within a completely flat structure. However, if there were to be a hierarchy in my organizations, productivity would be at the very top.
So, for example, here are a few random facts about me:
I don't hold any meetings and my calendar is clear
I have thousands of automations that assist me in my work
I've been working hand in hand with AI for over a year
I have one central dashboard where I manage literally everything
I've created my own AI clone that my co-workers can converse with
If this has piqued your interest, read on...
Day 0
Before the day ChatGPT came into existence, I didn't have much interaction with AI. However, for many years, I've been creating automations that assist me in my work and daily tasks. Most of my projects operate based on automation scenarios, which perform hundreds of thousands of tasks for me each month. These are the statistics from one of my many accounts on make.com, which processes 30-40 thousand operations on a good day. And that's just one account!
In this way, thousands of robots work for me, responsible for things like:
Handling my inboxes and triaging messages
Queuing posts for company social media
Adding customers to CRM and enriching their data
Handling subscription purchases
Onboarding sequences for users
Reminders for events, SMS, emails, and other notifications
And much, much more!
Essentially, every aspect of my businesses is surrounded by automations - from customer service, through back-office, sales, to marketing. This model of managing these processes allows me to run multiple projects simultaneously, with small, focused teams, and the robots work for me when I sleep.
Day 1
ChatGPT appears, which I treat as a curiosity, I check a few prompts, play with it for a while. It makes a pretty big impression on me, but as is usually the case, I don't initially see real applications and can't connect it to solving problems or improving my workflow. The answers aren't very precise and it's safe to say that it's not suitable for "production" applications. But over the next few weeks, I can't stop thinking about ChatGPT and further AI developments make me start devouring everything I can find on the internet about neural networks, advances in LLM development, and the future of what's coming.
The next big thing will start with something that looks like fun
This quote keeps ringing in my ears, even though I can't remember where I heard it.
Day 14
I am now convinced that in the future AI will change or significantly improve my productivity model. I feel like I'm about to take a big step, but I'm not yet completely sure how or when it will happen. On the other hand, I allow for a scenario in which all of this may actually stop and we will hit some kind of wall with AI. So I still approach the topic very cautiously, but I am putting together some patterns that I could potentially use in my work.
Not long after, it turns out that I'm not the only one who suspects that one of the most important applications of AI in the near future will be to increase productivity. Today, this is obvious to most leaders of the world's largest companies. Data from the PWC report indicate that it is estimated that the use of AI will give workers about a 40% boost in productivity, which is confirmed by Statista research:
Day 30
Aha moment! One simple thing that lifts the spell and allows me to demystify AI in terms of productivity. It turns out that LLM is quite good at transforming the commands it gets from me into the JSON format I use in all my automations. This is now common knowledge, used by many tools, but around the early days of ChatGPT, it wasn't yet obvious to practically anyone. This one simple thing (again, something that seems like fun) is the missing puzzle piece to connect AI with productivity. It allows me to run my automations in a completely new way - by having a conversation with the model. In this way, I can easily imagine a scenario where I or a person from my team gives a command to the AI:
Create an account for user X on our service for a month, then send the onboarding sequence.
This one simple concept allows me to imagine a completely different level of productivity, where the model is not so much able to connect with external services and interact with them. Rather, the model prepares data for my automation, which has been handling this task for a long time and does it excellently.
Now the only question remains: how the hell can this be done?
This is how the concept of Alice is born.
Day 60
After preliminary research, we find that no one else has come up with this or made a similar prototype, so we start working on an early version of Alice. The assumptions are simple:
A native application (because we are most productive at the computer)
Which allows interaction with LLM (via OpenAI API or other)
And the execution of our automations (with the help of sending formatted data to a Webhook)
During our work, these assumptions are repeatedly confirmed by others, as well as by OpenAI itself, which gradually begins to introduce Chat features similar to those we thought of for Alice. However, this does not change the fact that our concept is completely different, as it aims at a much wider range of typical productivity at work, which for me is associated, among other things, with very quick access to answers and the possibility of connecting prompts to keyboard shortcuts. This simple concept basically changes everything, and the solution that can help me has to be a native application.
After some time, we manage to create a working prototype that communicates with our automations and returns... complete nonsense 🙃
Day 90
After the first tests of Alice, my enthusiasm wanes significantly and just when I'm about to put Alice away, GPT-4 appears. It turns out that our concept may have been ahead of the model's capabilities, but... only for a moment. The progress in the field of AI is one of the most fascinating things I've seen in IT over the last 20 years.
It is precisely this pace of development that determines that you can currently start projects that are potentially useless, wait a while and... be a pioneer of a new, amazing market. That's how we think about Alice to this day, because although it gives us tremendous value, we will still have to wait a while for mass adoption of this concept. How long, is unknown - but I am almost certain that our concept will work out much better at the beginning than the assumptions of "general" agency, mentioned by OpenAI, which gets a lot of startups first excited, than extinct when OpenAI does it.
Why am I so sure?
The way we invented Alice doesn't immediately turn upside down what we're used to at work. The concept of agency in our version requires the preparation of automations that the model triggers. So it's not issuing a command into space, on the principle of:
book me a cottage in the country for the weekend
With all due respect, I dare say that creating such a universal tool at this point is completely pointless. It's an interesting theory and maybe one day we'll be able to get such actions from the model, but it also requires many infrastructural changes and software adaptation to the needs of such agents.
Unfortunately, I bet that even if such tools are created, the predictability and stability of the actions they perform will simply be unacceptable. Not to mention the fact that we are trying to change a paradigm here that may not be so bad at all? Maybe part of the process of resting in the country is choosing a nice cottage on AirBnb? Maybe people won't want AI to decide for them, because it's just a part of the process that they appreciate and that is enjoyable in itself.
It's a different story with the tasks we have to perform at work. Especially, with the help of interfaces that will simply be gradually replaced by AI and agency.
Why? Because they are simply ineffective. It's only the last few decades that have made us think that the mouse cursor or our thick thumbs on a small phone screen will be effective in issuing commands. As a result, performing a simple action often takes us a lot of time, when we have to fight our way through interfaces that are not adapted to devices, heavy interfaces, forget the login for the umpteenth time, fill in the same form with address data over and over again, and so on...
What was much closer and more tangible was what Alice was supposed to offer us. And we had no problem with the fact that it won't be (immediately) an application for the "masses" and will require a learning process. On the contrary - we were creating it for ourselves and our teams and the only goal at the beginning was to run our automations.
Meanwhile, Alice starts to perform simple operations, which we test, like this one.
Day 100
We have the first version of Alice, which works, executes commands, and triggers our key automations. It has the ability to add Snippets, which I configure under a keyboard shortcut. This allows me, with a shortcut and instantly, to perform, for example:
Correcting typos in any content
Creating a summary of the content I'm reading (e.g., an email)
Quick draft response to a message
Translating my Polish content into English
And much more...
Most of the Snippets that I spin in Alice are also assigned as shortcuts to Stream Deck, so I don't have to remember them all. It starts saving precious minutes during the day! The first people in our teams also start using Alice and begin to see value in it.
We also do the first experiments with external APIs, like Adam here.
Day 200
We're publishing the first page of Alice along with a waitlist, which over 3000 people sign up for. We begin the first application tests with beta testers and individuals we're training as part of our automation program. It quickly turns out that Alice can provide value not only to us but also to others!
Therefore, we're designing version 2.0 of Alice, which is to gain a target interface and usability, and perhaps be available to a larger number of people. We strive to design it so that it would also be the best and fastest possible way to interact with the model from the level of a desktop application. All this translates into efficiency and our productivity, and Alice can serve answers in record time.
I demonstrated how I work with Alice in a lecture for over 500 people, many of whom signed up for our lists:
Day 300
Along the way, we experiment with Alice's applications. When we explain Alice to people, it's hard for them to grasp concrete uses. In essence, you can do almost anything, but as we know, this isn't a tangible example. That's why we created a series of small examples of how Alice communicates with the applications we use, which I published on X.
We also prepared a business case in which Alice is able to register a customer in the CRM system and use a database of real estate listings, which it can match to them:
Day 365
It's hard to believe that our concept has evolved over a full year! We've released information about version 2.0 of Alice, which is meeting with increasing interest:
We're practically ready to make Alice available to a larger group of people, but the question is, are these people ready? We realize that Alice is not just a fast native interface for LLM and we don't want to advertise it that way. On the other hand, getting more out of it, like action handling, requires a basic understanding of automation techniques and, in general, our approach to productivity. It turns out that these early users who actually use Alice almost daily are people who have spent more time with us - participated in one of our programs or read our e-books.
At the same time, we already have a plan on how to develop Alice so that it can help an increasing number of people. However, this is not an easy task and we are beginning to realize that this might be a significantly bigger thing than we initially anticipated. On the other hand, this vision excites us tremendously and we already know that we would like to continue developing Alice not only for our teams. We've got a small taste of what we can call the future of work and although AI is not yet ready to serve it to us immediately, Alice is a great first step, where we get an ecosystem that can be better controlled, even though it requires more engagement.
From my perspective, productivity happens when we are able to adapt tools to our needs. None of the popular productivity frameworks or Notion templates, which offer to organize your entire life, have ever worked well for me. If we want to do it right, we need exactly what Alice gives us today - control. Unlike the promises of AI tools that will be able to do everything for us and currently fall into the category of utopian visions of their creators for me.
Today
Exactly one year from day 0, we are at a point where version 2.0 of Alice has already reached the first users. In our opinion, this is a version that is perfectly suitable for a wider audience, and currently the virtually unrivaled interface for LLM on the desktop. Interestingly, it probably would never have been created if we hadn't built it for ourselves. Gradually, it became so good that we also want to put it in your hands.
Here you can see how the Alice 2.0 application works:
Right now, our version of Alice has the following capabilities:
Conversations with different models like GPT-4, Perplexity, and Mistral
The ability to use models offline and have secure conversations with them
Generating images and recognizing them with Vision API
Snippets and keyboard shortcuts that can be assigned to them
Access to external services through remote actions and API
The ability to remember any information with a vector database
Voice control capability and reading responses with Eleven Labs
Generating images with Midjourney
The journey we've taken from day 0 is incredible!
Here, on the other hand, I've prepared a playlist for people who have a paid Techsistence subscription and have already received access to the application.
The Future of Productivity
In practice, however, what really resonates with me is the fact that Alice is not just a tool, but an entire way of thinking about productivity. After all, her features come directly from how we've developed certain work standards over the years, using an army of robots. And this is a completely unique approach that I haven't encountered anywhere else. Therefore, it can be said that the value lies not only in the application itself, but in everything around it, which includes:
Our set of tools and how we use them (Keyboard Maestro, Raycast, Shortcuts, and many others)
Our approach to time management and work
Hundreds of ready-made automation scenarios that we use
A functioning Second Brain in Notion and a knowledge base in Obsidian
And much more...
All these things are perfectly complemented by Alice and interaction with AI. That's why the people who discovered Alice with us use her daily to perform tasks, while many other people to whom we've made her available use her sporadically to interact with the model.
That's also why I have a specific question for you at the end:
We're considering preparing all these things in a form from which you could draw knowledge, as a 4-week program, in which we would guide you step by step through our methods, provide dozens of recordings, materials, content, and ready-made scenarios. We estimate that such a program would cost several hundred dollars and would also include 12 months of Alice application updates. Please let us know if you're interested - if we collect a minimum of 50 votes, we'll start preparations:
Please, let me know in the comments, what features of Alice could help you the most and how are you already using AI for productivity? 💬
Thanks for not being a stranger!
I’ve been following these posts on Alice for a while and am very interested. Currently I use Make.com a lot for managing the backend of a Wordpress e-commerce site. Also for sending dashboard-style notifications to team-only telegram groups to keep an eye on activity (sales etc). The OpenAI module is useful. But I haven’t been able to get much headway with personal automations. For those, I want self hosting so I was more interested in n8n, but it’s less user friendly. Very interested in Alice, it sounds something like raycast, but has an LLM input to its command set?
Setting up automations that are ready to receive JSON and then using AI to write those sounds super useful. I think what I’m looking for in something like Alice is to reduce the mental effort and time in setting up an automation. Being able to quickly go from idea to automation. Also one thing that stops me from setting up more automations is seeing how many edge cases there can be and how error handling adds a lot of overhead to maintaining a set of automations.