In 2024-2025, AI was all about prompt engineering and prompt tuning. That transformed into context engineering and agentic AI in 2025-2026! Now, I feel the next step is the Digital Twin! So what is a digital twin? Digital twin is one’s whole online and work persona “digitally cloned” online and running in the on-prem/cloud!

Basically, my digital twin for instance should be able to respond to queries from my friends and colleagues as me. It should be able to answer questions about my work and my life! It should be able to attend meetings as me and respond to emails as me! It should be able to remember information, understand my preferences, writing style, etc. and operate online given access to various resources that I have; arrive at a similar conclusion that I might in a particular situation.

Currently, instead of trying to sit on a beach and drink the coconut water, I want my digital twin to alteast automate my routine daily tasks. Everyday I spend some time going through my emails and responding as needed, prioritizing my activities/tasks, check on the bugs, project plans, etc. and adjust/add them to my priority list for the day as needed. This part can and should be offloaded to my digital twin. It would be amazing to open my computer and realize that my digital twin has already taken care of all this routines. Draft emails & responses are ready, possible updates to bugs and project plans are proposed and that my priority list of tasks is already ready for the day. I am sure there are various security considerations that needs to be taken into account before handing over such control to a digital twin, but those are solvable. I do believe that the next wave of AI is digital twins!

So, I have started exploring the idea. I did some experiments with OpenClaw but I still feel that it can do stuff, but it is just a framework and not a digital twin! It probably can learn my personality but it is costly (Peter S creator of openclaw burned 1.3M USD in a month), so it is prohibitively very expensive. Plus all the security holes and the verbose token usage does not make it a good choice. NemoClaw and PicoClaw are too based on similar concepts but NemoClaw needs permission fine-tuning and maintenance, and PicoClaw is not very powerful right now. So all in all, the claws are just claws and not persona.

To create my digital twin, I probably need a team of agents including a memory, persona, planning, research, drafting, task and automation agent (and a gatekeeper or safety agent to avoid and stop risky actions like formatting my hard-drive!). I will probably also need an orchestrator agent that manages all these other agents. I have been exploring various AI agents, frameworks, etc. that can help me with the objective. Local AI (Ollama/LocalAI), Open WebUI, Microsoft Agent Framework/CrewAI, some sort of graph memory and maybe an automation layer that provides tools to the various agents.

I still haven’t decided how to go about doing this. I have bits and parts working but not the complete flow. I will write more once I get there.

Happy AIing!