The next version of Argil will be available in just a few weeks, and with it a new vision that was born at the intersection of AI, workflows, and automation:
Argil’s AI assistant.
Essentially what all these new AI tools are providing people with, are ways to achieve tasks, better, faster, and cost efficiently but do we know just how big the automation market is?
Well, I didn’t so here are some numbers:
Quite big yeah, but since the launch of chatGPT and other foundational model these numbers no-longer reflects the real potential of the automation market.
As a matter of fact, here’s what people think about when you talk about automation:
Automation is seen as connecting tools to facilitate workflows, but that vision is limited to a predefined sets of functions. The automation market could only grow to the extend of manual/repetitive tasks available to automate.
No one saw it coming, if you rollback and ask people 2 years ago what would the software automation market enable they would have answered with another mundane task.
Anything in which human strategic and creative thinking was out of the equation…
This was true for a time, but the state of the automation market have changed, and here’s a list of tasks AI can already do:
The best about it, is that AI can respect instructions and processes to achieve these tasks. This means that the rigidity and inflexibility of existing automation tools can be solved by integrating AI.
I’ve been using AI tools daily for the past few months and I am now convinced that the potential of the automation market once AI is integrated efficiently will target not only the blue collar jobs but also the white-collar workspace.
Automation with AI will target tasks that require deep analysis, insight extraction, and decision-making.
It will provide detailed analysis of vast datasets to provide actionable insights quickly, and streamline these processes to create outputs in the required format to share with decision makers.
Analysis capabilities + complex problem solving + instructions = Automation revolution.
If you want to be one of the first to try the product, join the waiting list: Here
Based on this understanding, we decided to extend our vision of AI workflows to AI assistants.
Our AI workflows were at the intersection between the future outputs possible to generate with AI automation and the past due to the user interface and experience inherited from existing automation tools.
Our users didn’t bond with this type of interface, and the type of automation possible weren’t painful enough to get past that user experience barrier.
On top of that we reflected on what was the current best example of concrete AI tool adopted at scale.
What OpenAI did with ChatGPT was the perfect use case.
So we identified 5 pains associated with it’s usage:
These learnings were the initiator of the shift, AI workflows needed more intelligence, and we found that intelligence through a user experience based on AI assistants.
LLMs emergent abilities showed they excel at following instructions.
On top of the unlimited realm of applicability of models (datasets they’ve been trained on), you got the perfect mix to build personalized user experiences at scale.
Our approach to AI assistants is based on 3 pillars:
Here's a little preview with a Content Creation AI assistant:
We see the LLM as the executive layer of human intuition, enabling hypothesis A/B testing at scale to find the framework that emulates the best the intended output.
Instructions are how you get a SOP (standard operating procedure) approach to how the LLM achieves a task.
It’s the representation of the framework to follow to get the output you want:
This became possible with the inherent capabilities of LLMs to predict.
Next token prediction (Next-token predictors are models that predict the most likely word or symbol to follow a given text sequence.) allows anyone that sets the assistant the first time to get predictability when they re-use it for their intended purpose.
If you want to be one of the first to try the product, join the waiting list: Here
After pitching the current version of Argil to 50+ people, half of them asked the same question:
‘Can I make multiple assistants work on a task?’
That’s what happens when companies start to pitch to their users what they pitch to their investors. Present expected value become what would be possible in a near future (2-4 years). You break the user experience by selling them the future.
There are limites now to the autonomous vision, I know this might break some hearts but what you’re seing on some demos is fake, and wont be possible before a few years.
LLMs are stochastic, they can hallucinate or as I rather say, be creative.
This means that in a SOP with multiple agents interaction, you increase the probability of hallucination and decrease the accuracy of the output you’re getting.
Our approach solves that with through UX:
A step in the present, and one in the future.
Here’s what you can already do with our AI assistants:
Ok this is good Othmane but what exactly are Argil AI assistant?
See them as your virtual companion designed to centralize and outsource some of your daily tasks. They represent the embodiment of specific workflows and processes you go through.
We’ve built them in a way to represent at best your ability to create respecting your instruction and your approach to task achievement.
Our AI assistants are about bringing together effort, context, and skills (but you’ll get what I mean by that in a second.)
Here’s what is meant by tasks:
If you want to be one of the first to try the product, join the waiting list: Here
1/ AI assistant tasks:
A task is essentially a standard operating procedure that’s aimed at delivering the specific output you want following the framework you’ve instructed based on the inputs you’ve set.
A task is the current best way we came with to use LLMs at their best capacity blending them with your directions on how and what to use to get the job done.
Each task is composed of:
2/ AI assistant skills:
Skills are the innate capabilities that you can link to each of your AI assistant tasks, as each task will use different input of content or source of triggers they need to associated skill to make start the SOP of the task.
Essentially, skills are your key to using the input you want to get the output you want.
At the moment, the skills include the ability to interact with a PDF or a webpage, write in a certain style, and generate images but that’s just the start.
Last, but not least.
Work is collaborative, and the current approach of AI tools is individualist.
This creates a real scission, which is why we thought the experience you can get with our AI assistant to be collaborative with a Workspace approach.
3/ AI assistant and workspaces:
One of the main limitations of chatGPT is the discussion based approach which highly impacts the collaborative potential of the tools. How can I centralize the discussion related to a specific project i am currently working on? Not possible.
Workspace on Argil allows you to do this and collaborate with your team on the work at hand with the stakeholders.
Centralization of task management + AI Assistant tasks + History of work = Massive gain in productivity.
Our AI assistant will boost team work efficiency, while some are calling AI to take people jobs, we’re focused at redefining what work will mean by providing the best collaborative experience.
If you want to be one of the first to try the product, join the waiting list: Here
If you have an urgent use case and want to know if it’ll be possible on Argil, send me an email at othmane.naimkhadri@argil.ai
