Prompt engineering started as a quick fix for a common glitch in the large language models (LLMs) driving the latest wave of artificial intelligence and machine learning. Crafting instructions to guide GenAI applications has opened a world of where humans and computers align to achieve desired output, including the generation of research or a string of commands that tell AI agents what actions to take in any given situation. But just as the role of prompt engineering jumped on the scene to become a must-have skill, it jettisone ahead into something that may be ever-changing, according to Daemon Behr, Field Chief Technology Officer for the Americas with Nutanix.
At every blink of an eye, prompt engineering seems to be morphing.
“In mid-2025, the industry collectively changed all mentions of prompt engineering to context engineering, although most of the same strategies and processes apply,” Behr told The Forecast.
He said ideas around in-model memory, and out-of-model memory.
“In-model is managed by reasoning models and is more ephemeral, limited to a conversation, with some lossiness over time. The longer a conversation, the more it forgets and needs to be reminded.
“Out-of-model memory is when agents create artifacts that are exported after key tasks, and then use as future context. This helps with alignment of outcomes, and reduces hallucinations and lossiness of context. This can fully be created by context engineering, and does not rely on model specific capabilities.”
Behr is one of the people helping Nutanix fine-tune generative AI, or GenAI, for its cloud-services customers. Prompt engineering is increasingly pivotal to delivering these kinds of AI-powered services. Behr talked The Foreacast about where prompt engineering came from, where it’s going now and where it’s likely to be in the future.
Humans won’t use prompt engineering to hand off their thinking to machines, he said. Rather, they’ll use it to improve their thinking by collaborating with AI models. “If you allow AI to do your thinking for you, you're going to get a middle-of-the-road, generic result,” Behr told The Forecast. “If you do the thinking in collaboration with AI, then you're going to be able to accelerate your capabilities. If you don't use AI, you're not even going to be on the board.”
LLMs Light a Fire for Prompt Engineering
LLMs launched a new computing era in late 2022 with the announcement of ChatGPT, which allowed users to enter plain-language prompts in an online interface and get text or other content that seemed almost uncannily human. For all its merits, conversational AI has an inconvenient flaw: Computers don’t always know what they’re talking about.
“The models hallucinated so much that in order to get any kind of quality result, you had to really focus on prompt engineering so you could ask very, very specific questions to get very, very specific outcomes,” Behr said.
Behr has been scrutinizing the progress of ChatGPT and its many competitors in the rapidly changing GenAI landscape. “As the models have become better, prompt engineering has changed in focus from preventing hallucinations to doing more interesting things,” he said. One of his hobbies has been helping Nutanix staffers see the potential of prompt engineering.
Behr built on this idea in his blog and podcast. His goal: Use GenAI to boil vast amounts of information down into a book-length document that would reveal connections between LLMs and specific knowledge domains. Prompt engineering guided GenAI apps through the process of writing the text.
One book converged the possibilities of GenAI with Kaizen, the Japanese concept of constant improvement, which Toyota employed to transform the automotive industry. The other merged GenAI with Fika, a Swedish concept designed to help people find a flow in life. “Fika teaches us to pause, to find balance, and to embrace moments of mindfulness amid the rush of daily life,” his book stated.
Publishing books on these topics with conventional tools would’ve taken an author months to research and synthetize into a completed book. With prompt engineering and available GenAI tools, Behr created them in a few days.
Behr didn’t intend GenAI to replace the work of an author. Rather, Behr saw it as a venue for curious people trying to navigate expansive ideas without drowning in data. “It would be like having a boat with a motor in the river of information,” Behr said on his podcast. “This is the boat that I'm building with generative AI.”
How Prompt Engineering is Evolving Now
Behr is mastering prompt engineering against Nutanix’s backdrop of putting GenAI to work for its hybrid cloud-services customers. The company built chat-based GenAI interfaces to help salespeople, systems engineers and customer support staff quickly find and summarize complex technical materials, for instance. Prompt engineering lets Nutanix staff craft increasingly sophisticated queries.
LLM frameworks come in many flavors. Some allow chain-of-thought prompting, where each response lays the groundwork for further prompts. “It becomes very, very refined so that the actual response it gives you is much, much more focused than just giving it a single-shot prompt,” Behr said.
Others entail building large things like knowledge bases based on internal company documents, while others analyze large datasets and deliver concise summaries. Either way, prompt engineering lets users drill down to where they need to be.
Software developers also can confer with an LLM on code design. “Let's say you have a large project and then you communicate with the model and you say, ‘this is what I'm trying to achieve — build out a software spec,’ and then that spec would be broken now into different components or modules, and then they could go in and interactively talk via natural language to build those specific modules,” Behr said.
Indeed, an LLM can help people work faster and more effectively on just about any task with a language component. Users just have to remain aware of how LLMs operate. It’s easy for an LLM to answer simple questions via ChatGPT or Perplexity. Developers can use API calls to pull this basic functionality into applications.
“If you do that interactively in short bursts, then that's fine. It can provide some great quality,” Behr said. However, developers must be careful when, for example, analyzing an entire code base or attempting other compute-intensive prompts. “That gets extremely expensive because you're paying for every single API call,” Behr cautioned.
Where Prompt Engineering is Going
GenAI interfaces connect to intelligent software agents that users never see. These agents represent the future of prompt engineering, Behr said. In software development, for instance, one set of prompts can instruct an agent to analyze an enterprise’s legacy code base so that it can be refactored in a modern language. Another agent can analyze the new code to ensure it follows best practices for security and usability.
New standards and protocols like A2A (Agent-to-Agent) and MCP (Model Context Protocol) are speeding the evolution of prompt engineering. “You're going to be communicating with a mesh of agents that have different capabilities,” Behr added. “It's like an entirely new way of accessing knowledge and that knowledge is going to keep on getting generated and added in real time, but it's not going to be done in the same way it has in the past.”
This is welcome news for techies building the software platforms of the future that will keep all the agents talking to each other. But it’s also a warning that people will need training in making sense of this new way of computing and communicating.
“It's going to become a requirement to get any type of job,” Behr said. “The quicker that people are able to understand that and get a base amount of knowledge, the more they're going to be able to keep the jobs that they have and have greater opportunity.”
Editor’s note: Learn more from Daemon Behr in the book Designing Risk In IT Infrastructure and his Canadian Cybersecurity podcast.
© 2026 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.
Tom Mangan is a contributing writer. He is a veteran B2B technology writer and editor, specializing in cloud computing and digital transformation. Contact him on his website or LinkedIn.