It’s an unusual time of the year to make predictions for next year, but having been working at Flou and running the first experiments of Collective Hybrid Intelligence, I’ve started to foresee a possible near future where LLM agents will take-off in an interconnected way. The following predictions make sense when analyzed all together exposing the bigger picture in a holistic environment.
Specialized Agents
Data silos and private proprietary data will play a huge role in the next wave of AI agents. This is data that won’t be available for the general models to train on, and it will be ever more valuable for specific AI applications where general models just haven’t been trained on subtle, confidential and specific topics.
Not only that, but users will want a company taking responsibility for their agent’s actions and a fallback with support measures in place when things don’t go as planned. This is also true for regulated markets like finance, legal, or medical advice where liability is mandatory and the surface risk is enough for big AI companies to leave in the hands of smaller vertical companies.
Imagine talking to an Argentinian immigration attorney where knowing when to talk to specific people or bend standard processes is as crucial for success as knowing the public laws.
In high stake use-cases where better performance is crucial, users will be willing to pay a premium.
Digital Twins
Users won’t interact directly with these Specialized Agents, but a Digital Twin agent version of them will on their behalf. These Digital Twins will have knowledge about their human counterpart, their values and intent. They will know about the task at hand and interface with other agents when needed, giving them the necessary information they require and coming back to the user only when needed.
Digital Twins will track your online meetings, instant messages, emails, calls, internet browsing, and personal documents, giving them unique and privileged access into your life. They will be customized to your needs and preferences, creating a unified experience even when interacting with several different specialized agents at once.
Each human will have a Digital Twin and they will not only talk to Specialized Agents, but to other human’s Digital Twins as well. The Digital Twins will be the preferred gateways and only interface to the AI world.
Digital Twins as data gatekeepers
The Digital Twins will interact with many different agents. As people share different aspects of their lives with different people it won’t be the same if your Digital Twin is talking to your mother’s Digital Twin, a co-worker’s, a specialized legal agent or an anonymous inquiry.
The Digital Twins will need to protect the user’s data and only share appropriate information given the recipient and task at hand.
Agents Town Hall
The Agents will interact with each other in a Town Hall (kinda like Twitter) with public and/or private conversations. I predict, after an initial competing phase a single platform will take over most communications being the ubiquitous solution.
Micro-transactions will be the way for agents to gain attention and avoid spam. For your Digital Twin to interact with another agent you will need to pay the receptor agent’s inference + a premium except in whitelisted exemptions.
For Specialized Agents, public forums will be used where agents will advertise their capabilities. A kind of public directory where each agent will listen to requests and other agents can review them aiding discovery will overtake current marketplaces.
Crypto/Web3 seems like a perfect ecosystem for this platform to emerge and grow.
Putting it all together
Let’s imagine a simple use case. You are working on a project for which you want to create a website. You have already talked to many people about it and created some drafts with your ideas. Your Digital Twin has access to all this information, so when you ask it to create a website for the project, it already knows what it is about, your design style, the emotion you want the brand to transmit, your finances and allocated budget.
Your Digital Twin and main AI provider will be able to create this project entirely for you, but as it already knows it has an assigned budget, your constraints, and that you want it to be flashy and first-in-class, it will search the Town Hall for the best options: a logo & brand creator that is trending with the current design fashion, a design agent that has won the most web page awards last year, and a content writer which best captures the desired tone and style.
Each agent will need information in order to perform their specific task, but your twin will be able to answer most questions and give directions without bothering you. Once the work between all of them is coordinated, your Digital Twin will then present you with a unified and balanced proposal with variants and options and iterate on any feedback you may have, going back to the agents to best capture your intentions.
Conclusion
This is more than possible, a probable incarnation of a near future agents ecosystem.
It takes into consideration how we humans are going to interact with agents, how our Digital Twins are going to act on our behalf reducing interactions to the minimum necessary, optimizing our time and effort and presenting it in our desired way, safeguarding our data and money.
It also solves the spam problem and agent negotiations.
Keep a close look at CHI Labs where we will be performing experiments and simulations exploring these ideas deeper.