Techie Girl

Write "The team is making great..." and your mind already expects an ending. A language model gets there through a sequence of calculations. You draw on experience with language. The model uses patterns learned during training to work out what could follow. Microblog @antgrasso A computer needs numerical representations to process language, but assigning a number to a word is only the beginning. The model divides text into tokens, which may be whole words or smaller pieces. Each token receives a numerical ID, essentially a label that points to a learned vector of numbers called an embedding. Now comes the interesting part. As the model processes these vectors, it builds representations influenced by the preceding context. The same word can contribute differently depending on the sentence. After "The team is making great...", the model might assign a higher probability to "progress" than to "decisions" or "plans". It then selects a token and uses it as part of the context for predicting the next one. And the calculation continues. Change the preceding text, and those probabilities may change too. Even a small modification to your prompt can influence what the model generates. #LanguageModels #AI

After learning how RAG gives AI better context, the next question is how AI connects to the world around it. MCP, or Model Context Protocol, helps applications reach tools and data in a structured way, while governance defines what they can actually do. Microblog by @antgrasso RAG, or Retrieval-Augmented Generation, helped many people understand that AI needs more than a model. It needs relevant knowledge from approved sources, so the answer is grounded in the right context instead of relying only on what the model already knows. MCP moves the conversation one step further. If RAG helps AI bring knowledge into the answer, Model Context Protocol helps AI applications connect with external capabilities in a more consistent way. That connection matters because useful AI work rarely stops at generating text. An application may need to search records, check information, update a ticket, or support an operational task. MCP gives this kind of interaction a clearer structure. Governance is what keeps these connections safe. Governance is what keeps these connections safe. Permissions and usage limits define what an AI application can read or activate, so connection does not turn into uncontrolled access. In the end, RAG gives AI better context, while MCP gives AI a more structured way to act. The value comes when both are designed with clear boundaries from the start.

A technology project can succeed technically and still deliver little value if the way people work does not change. I believe organizations should plan those changes as part of the investment, not after the technology has been deployed.

An AI agent can initiate blockchain transactions and perform DAO operations, provided it has an authorized wallet and the required permissions. An agent can go beyond recommending an action and carry it out. People still decide the limits of that authority. Microblog @antgrasso Working with a blockchain is different from updating a conventional database. An application with database access can change records, but it cannot simply change a blockchain's shared state. Transactions must meet the rules checked by the network. The AI agent itself usually runs off-chain. It analyzes information and decides which action to request, while the blockchain validates submitted transactions and executes any relevant smart-contract code. The AI model does not perform its reasoning on the blockchain. A decentralized autonomous organization (DAO) allows a group to coordinate decisions and manage shared resources through blockchain-based governance. Despite its name, a DAO does not need AI to operate. But it can delegate specific rights or tasks to an AI agent. An agent could therefore cast a vote or initiate a treasury payment if the DAO's rules permit it. This does not mean giving the agent unlimited authority. The organization can define which actions are allowed and which still require approval from its members. Smart contracts can enforce programmed limits, although they cannot determine whether the agent's judgment is sound. The agent can act autonomously within its assigned role, without taking control of the organization's governance. #AIAgents #Blockchain #DAO

ADIPEC 2026 is almost here, and the world's largest energy event is growing again. From 2–5 November at ADNEC Centre, Abu Dhabi: 2,260 exhibitors, more exhibition space and 1,800 speakers. More opportunities to connect. Join > adipec.com @ADIPECOfficial Partner. An expanded exhibition does more than welcome more visitors. It allows companies to show what they are building and gives potential customers more solutions to examine in person. For me, that is what makes a major industry event useful: a direct conversation can reveal whether a technology fits a real business need. Bigger in scale, bolder in ambition, greater in impact. A bigger, bolder, better ADIPEC experience. #ADIPEC2026 #ADIPEC #Energy