January - the time of resolutions and predictions. Personally, I predict I'll have already broken my resolutions before you finish reading this.
An easy prediction to make? You’ll love our weekly newsletter. An easy resolution to keep? Subscribe and hit reply to tell us what you'd like to see from us in 2025.
Speaking of predictions, we asked some members of the community to share their thoughts for the year ahead.
Let's see which predictions will take off like GPU costs, and which will crash like an unoptimized model on production data… |
|
|
| Senior Software Engineer at Bloomberg LP | Podcast Host |
|
|
Which tool, framework, or company will dominate MLOps in 2025?
2023 was all about the hype around generative AI. In 2024, we’ve seen an initial wave of Gen AI applications that are starting to introduce real productivity gains within the enterprise. 2025 will be all about optimizing the underlying infrastructure to help scale these solutions.
To that end, the focus for tooling will be on optimizing costs and efficiency of generative AI models. Working at Bloomberg has enabled me to get directly involved with the open source tools and infrastructure that many organizations are using to deploy their models to production across the enterprise. The recently announced Envoy AI Gateway is an exciting piece of infrastructure whose features will help companies optimize model costs through autoscaling based on token usage and implement request prioritization. Improved GPU utilization, better LLM storage solutions, and prompt caching features in deployment tools like KServe will continue to make it more accessible to deploy a variety of models.
We’ll also see a big surge in the use of buzzword-heavy AI concepts like Retrieval-Augmented Generation (RAG) systems, generative AI, and cloud-based AI products, all of which will become easier to use and, hopefully, cheaper, thereby driving further broad adoption.
High-quality data will continue to take center stage. As AI becomes more integral to decision-making workflows, companies developing AI products will be held accountable for the outputs generated by their AI models. I believe this will drive investments in explainability and data quality. This is especially true in the healthcare field, where making a wrong suggestion to a user could negatively impact their health. What’s the tech, company, or trend that will fail to deliver next year?
I think we’re likely to see some backlash against overhyped buzzword-driven trends in AI. There’s plenty of exciting innovation happening outside the “trend of the month” spotlight that deserves more attention, particularly in areas like quantum computing.
Who or what should we keep an eye on in 2025?
I’m excited for industrial and space tech developments that use AI. These two sectors are pushing the boundaries of what’s possible by using AI in innovative ways. For example, AI is being used in manufacturing to optimize assembly processes, automate complex workflows, and enhance precision with tools like 3D rendering and autonomous robotics. These advancements are making production faster and more efficient. I believe companies like Dirac Inc., Atomic Industries, and Solideon are revolutionizing this industry.
In the aerospace and space industries, AI is enabling breakthroughs like microgravity manufacturing, autonomous inspection systems, and the creation of next-generation transportation technologies. Similarly, the maritime and automotive sectors are seeing rapid advancements with AI-powered autonomous vehicles and sustainable transport solutions. In this space, I think companies like Varda Space Industries, REGENT Craft, and Saronic are driving innovation with their AI-powered vehicles.
One thing you’d love to see happen:
I’d love to see more advancements in and attention around space, industrial, and healthcare tech that pair AI with hardware innovation. Physical products that integrate AI to drive real-world impact are always inspiring, and I hope 2025 brings more breakthroughs in these fields.
Boldest prediction about the future of MLOps that might sting come Dec 2025:
AI presents incredible opportunities for discovery and advancement, but we should be careful that it doesn’t come at the cost of human connections. As we adopt AI more broadly, we must be mindful of how it affects social dynamics. I think AI will create the illusion of connection, but if we try to replace it with human connection, we as a society will become more isolated and lonely. I think we need a societal shift in how we view AI, how easily we give trust to it, and what we rely on it for. I don’t think we are dedicating enough effort to these questions, and that lack of reflection may come back to bite us down the road, especially since this technology is developing at such a rapid pace.
I recently had a great discussion on my “Alexa’s Input (AI)” podcast about this topic with Devansh, author of the popular Artificial Intelligence Made Simple newsletter on Substack. Listen to it here. And I’m hoping to talk about some of these other subjects discussed above with guests in upcoming episodes.
|
|
|
|
Staff Tech Lead, LLMs @ EvolutionIQ |
|
|
Which tool, framework, or company will dominate MLOps in 2025?
Modal What’s the tech, company, or trend that will fail to deliver next year?
Langchain, horizontal "LLMops" companies. Maybe not 2025 but eventually. Who or what should we keep an eye on in 2025?
Loads of small 1-5 person companies that are supercharged in productivity through LLMs and build niche, profitable companies that would have otherwise required dozens of engineers, and not funded by VCs. I also think Anthropic will continue to grow and might start to dominate chat users, while Gemini starts to dominate the API user base for tasks. One thing you’d love to see happen: More introspection and explainability. More of the "mapping the mind of the LLM" from Anthropic. Boldest prediction about the future of MLOps that might sting come Dec 2025:
1 - People will use LLMs for simpler tasks, not more complex ones. The complex ones will largely flop because people went too big too fast and lost their first-principles product work. They'll step back and start creating more foundational LLM building blocks. Only then will they start to grow in complexity. Walk before run.
2 - Really good PMs and engineers will actually start to converge. With LLMs, coding won't be enough to differentiate as an engineer, you'll need to think about the product, business KPIs, strategy etc. You need to think about solutions to problems, not software tools. And PMs are going to be expected to get more technical. Nothing is stopping them now, LLMs help you code. They will need to use their product knowledge and combine it with technicals to give engineers more tangible instructions. Those sound pretty similar, and the result will be less fragmentation and miscommunication between PMs and engineers, something that's far too common today. They'll start speaking more similar languages.
|
|
|
|
AI Consultant @ Sentick · Freelance | |
|
Which tool, framework, or company will dominate MLOps in 2025?
Predicting a single dominant MLOps leader in 2025 would be hard so I am going to go with two:
1 - The MCP standard that Anthropic introduced, or something similar to that, to make it faster and easier to connect models to sources to accomplish a task. This would enable models to do tasks involving multiple applications more reliably.
2 - Memory management for agentic systems: Such systems would help AI agents retain and leverage richer, more personalized context over longer periods. By effectively "remembering" past interactions and adapting to user-specific preferences, these tools would improve personalization and overall user experience. What’s the tech, company, or trend that will fail to deliver next year? Startups and companies that primarily focus on creating "wrappers" around existing AI models—without offering real value—are likely to struggle. Who or what should we keep an eye on in 2025?
Watch for the rise of reliable, agent-like AI systems capable of handling more intricate, multi-step tasks. This could mean they'll take over repetitive, time-consuming parts of knowledge work.
One thing you’d love to see happen: More experimentation in "computer use", much like what Anthropic is doing. For example, a general vision-based navigation tool could let an AI handle everyday errands, like booking a hair salon appointment, without you having to micromanage the process. Boldest prediction about the future of MLOps that might sting come Dec 2025.
Most knowledge workers will be using AI to automate at least some parts of their lives. I'm hopeful that these tools will be so user-friendly that even non-technical people can easily pick them up and integrate them into their daily routines. |
|
|
| Experienced data and AI leader @ Wicked Data |
|
|
Which tool, framework, or company will dominate MLOps in 2025?
This is a really hard one! I am a bit tied between Google, Meta, and Anthropic at the moment. All 3 are making big strides in different areas: Google has the best multi-modal support, Meta has the best "open" systems, and Anthropic has one of the most "loved" products right now.
If I really had to choose one, I'm betting on Google. I don't think people actually care much about AI systems being "open" -- that will get Meta good "press", but if the service is cheap enough, people would much rather just pipe their data to Google (which they probably already have anyway).
The deep integrations into the gsuite ecosystem gives them a market advantage, and they are KILLING it with multimodal support. What’s the tech, company, or trend that will fail to deliver next year? I think we are starting to see the end of vector/embedding storage companies, personally. The reality is the problem they solve is too much of a "sledge hammer".
There's a common vibe in ML where we throw away all the lessons of data management systems, and then it comes back to bite. This is another such case. We have been doing retrieval for decades, often with embedding based methods too! The tools we have (e.g. traditional search indexes) are certainly not perfect for it. But the vector/embedding storage solutions generally throw away all the other things we have learned. They are easy to get you to a prototype/started (which makes them easy to buy/adopt for companies looking to get started quickly). But they eventually become a bottleneck in delivery. As companies develop better/smarter AI practices, vector/embedding storage will get squeezed out. Who or what should we keep an eye on in 2025?
I am currently most conceptually excited about what the people at Mirascope are cooking. My entire career working in AI/ML has shown me that domain knowledge is the highest leverage investment. Most of the tooling for AI/agents/etc really is still in the "technical" realm, intended for AI/ML people or maybe engineers. But imagine a world where the domain expert had the agency to iterate and improve an AI, without having to go through an "AI engineer intermediary". That would open the door to create incredibly powerful AI systems quickly. Keep a close eye on the product they are building, because I think it will close this gap.
One thing you’d love to see happen:
Now that we are starting to see a lot of movement in integrations/tools for agents, I would love to take that a step further with curation and connection. Much of the challenges we have with AI/ML efforts before now has largely been data related. Can we get the right data to the right place with the right quality? In practice, this has been difficult. But AI offers us some potential solutions/tools that could help us curate and connect our data to make it maximally useful for AI solutions.
Boldest prediction about the future of MLOps that might sting come Dec 2025:
Solopreneur AI practices will skyrocket as companies seek to optimize cost structure in an uncertain economy (esp with economic policies proposed by incoming President Trump). This will lead to slow hiring rates but still high desire to adopt AI practices. But after > fiscal year with lots of AI investment across the board, and little to show for it, companies will either give up the dream entirely or at least be much more discerning about how it affects the bottom line. Hiring AI consultants will be the "least risk" option here.
|
|
|
|
Chief Happiness Engineer @ MLOps Community |
|
|
Which tool, framework, or company will dominate MLOps in 2025?
Agents, agents, agents. What’s the tech, company, or trend that will fail to deliver next year?
Adept AI, Character AI, and Inflection AI. Who or what should we keep an eye on in 2025?
All of the MLOps Community local organizers. One thing you’d love to see happen:
An in-person Agents in Production mega meetup in NYC in May (oddly specific because we are currently looking to close on a location and make it a reality! Block your calendars!)
Boldest prediction about the future of MLOps that might sting come Dec 2025: Mistral founders leave the company and go back to work at Facebook. |
|
|
So that’s our thoughts. Got a different take, or feel we’ve missed something? Reach out on X and LinkedIn and let us know!
|
|
|
|
