The Age of Agentic AI:

The mind boggling speed of technological evolution is only going to accelerate.? As we start to approach technological singularity or artificial super intelligence (ASI) , an AI system that surpasses human intelligence in every domain, capabilities will evolve that are unimaginable to humans today.? If you think this is science fiction then strap on your seat belts and get ready for the ride of your life.? We got a preview of rapid technological acceleration this week when the hottest idea of last year (Large Language Models - LLMs) became commoditized by the introduction of the open source DeepSeek R1 model.

Until very recently, it was a general belief that to compete in the AI field companies needed investments in the range of hundreds of billions of dollars. DeepSeek shows that good old fashioned creative thinking can spin that paradigm on it’s head.? DeepSeek leverages mixed precision, chain of thought, reinforcement learning and model distillation concepts to beat the performance of some of the best LLMs in the market at a fraction of the cost (benchmarks such as AIME, MATH-500 and SWE-bench).? Suffice to say DeepSeek is “real” and for those critics that point to the fact the DeepSeek is leveraging capabilities built by other (well-funded) competitors - all I can say is that we stand on the shoulders of giants.

LLMs are trained to predict the next token in a text sequence and generate responses based on learned patterns in their training data.? This makes them a powerful tool for search, but similar to google searches LLMs require continuous prompting to get answers.? AI Agents in contrast are designed to be action-oriented.? The commoditization of LLMs has opened the floodgates for action-oriented AI.? Agentic AI leverages LLMs as part of its reasoning component, but it adds layers for action selection and environmental feedback loops.? Feedback loops typically incorporate observation, decision and action.? This means Agentic AI requires integration of real-time data feeds and APIs to interact with external systems.? This structure lets the AI refine its strategy as it encounters new conditions.

In the early 1990s I had the luck of working with the late Steve Brosky (and others) on NASA’s CORE project that required guaranteed response time from the operating system.? This was my initiation into mission critical computing where timely collection of sensory data from thousands of input devices was essential to mission success.? Imagine an AgenticAI system that has reactionary capabilities at sub-microsecond latencies, it could help humans avoid disasters like the one experienced this week in DC with the American Airlines flight.? My heart goes out to the the victims’ families.

In order for us to collect and store data from thousands, if not millions, of devices in real time the storage architecture becomes pivotal.? I am a huge fan of cloud native object storage capabilities that enable hybrid cloud deployments (ala MinIO / AWS S3).? However, there are thousands of applications (think about sensory devices) that have been written using POSIX semantics (classic file systems).? Action based AI will require that we collect data from every resource and make it available for analysis in real-time.? There are not that many storage products that have been time tested in mission critical environments to deliver real-time intelligence, think nuclear testing.? Some of the largest super computers on the planet running at the department of energy or defense (DOE / DOD) leverage machines that deliver high bandwidth data movement and reliable storage.? Incorporating these real-time storage systems with AgenticAI workloads will proliferate our lives.? I think that if Nvidia’s DGX SuperPOD and DOE’s supercomputers are deploying DDN’s data intelligence platforms (https://www.ddn.com/products/a3i-accelerated-any-scale-ai/) then the question becomes how can we incorporate similar capabilities into stand alone AI agents?? Fit for purpose AI will become visible in every aspect of our lives.? We will need more GPUs, CPUs and storage - not less.? Artificial super intelligence when available will explode the availability of AgenticAI systems in ways that none of us can imagine today.? We are blessed to be living in these interesting times:) .

To be certain, data security and ethics will need to be carefully considered as these systems evolve.? For instance if autonomous decision-making could help save human lives, who bears the responsibility if an AI agent makes a wrong call?? Similarly, how can we ensure that data privacy and security remains front and center as these agents leverage information for general usage?? These are deep topics, and I may explore them in future posts.

#llm #ai #artificialintelligence #ddn #bigdata #cloudcomputing #innovation