Insights from Miami AI Agent Summit 2025
- Cristian Dordea
- Apr 8
- 7 min read
Last Saturday, April 5th, I attended the Miami AI Agent Summit. There were over 250 attendees, and they had five amazing back-to-back panels for a full day centered around AI agents.
The CrewAI CEO kicked off the event, followed by the Hugging Face CEO then by 8 others including the Lenovo CTO, VC investors and many other AI start-up founders.
These founders and discussions uncovered many lessons and insights into the AI Agents arena and the opportunities as well as challenges of the AI start-ups in this space.
In this article, I will recap the main takeaways from this event, highlighted by each of the 5 panels. Make sure you don't miss my own main observations and advice that I found the most interesting at the end of this article.
Panels
João (Joe) Moura - Founder and CEO at CrewAI
CrewAI is taking off and working behind the scenes with large enterprise clients.
They are realizing a AI Agent framework is not enough at scale when you have 1000's of agents running. An enterprise suite is needed for all the Agentic resources.
As one of the fastest growing multi-agent platform, they realized the importance of Interoperability, Observability and Governance once you have scale with AI Agents
Use cases are expanding (see photo below on use cases by top verticals or top functional areas.)
Some of the new things they are announcing:
CrewAI job board listing jobs around the world requiring CrewAI skills
Clem Delangue - Co-Founder and CEO at Hugging Face
He is excited about the AI agents being integrated into robotics solutions (he gave a fun example like the use of a robotic so100 arm)
He encouraged people to consider looking into robotics toys for kids and how the new generation toys can be enhanced with AI Agents. He sees a lot of potential in this space for the application of AI agents.
Transformer architecture will be here for a while and will continue to do so.
He emphasized the importance of Evaluation Frameworks for AI agents and suggested that 30% to 50% of the builder's time should be spent on the evaluation of the AI Agent.
He emphasized the need for AI Agents to be decentralized and open systems as well as transparency of the models to gain the trust of the consumer and, ultimately, the adoption at scale for AI agents.
He sees the primary risk the the AI models that don't have enough transparency and openness, especially if we are considering giving them autonomy in the future.
Customers and users don't trust AI agents for critical use cases if they don't know what the models do behind the scenes and what data they were trained on.
Garret Rowe Founder and CTO NeuralSeek
One of the biggest challenges seems to be integrating these AI Agent solutions with legacy enterprise systems.
Another challenge, as confirmed by CrewAI CEO earlier, is the orchestration and visibility of those 1000s of AI Agents at scale.
As the Hugging Face CEO also mentioned, traceability and visibility of what the AI agents do behind the scenes are very important for customer trust.
They noticed that once a specific domain problem has been identified and well understood, they are better off using small, specific, highly accurate AI models than large ones.
There was a range of opinions on what constitutes a company's moat in this space, but the majority seem to believe that the main moat is internal organizational proprietary data, deep knowledge of their own business workflows and processes, and expertise in their specific domain.
Hernan Londono CTO & Innovation Strategist at Lenovo
Ed Sim Partner at Boldstart
Ed Sim emphasized the importance of rebuilding the tech stack in the future so that the entire tech infrastructure, end-to-end, is AI Agent native once AI agents scale and become mainstream. He acknowledges that they are making bets in this space in terms of investing.
Hernan Londono sees a big opportunity at the intersection of IoT + Edge + AI Agent due to the low latency benefit and the advancement in the current AI models.
Ed Sim, a VC Partner at Boldstart, mentioned that he was one of the first investors in CrewAI. He also confirmed investments in Protect AI, a LLM Security company, as well as investment in a start-up tackling cloud infrastructure for AI agents
In terms of competition and staying ahead in this fast-paced, evolving space, Ed Sim emphasized the importance of speed of delivery, distribution, and owning one's own data and user base.
There is a mix of opinions on what the new UI would be for AI agents.
Another big question that remains to be answered is how authentication and identity will be handled for AI agents once we give them autonomy within an enterprise.
They both acknowledge that we are very early with AI Agents, and there are many points of view on where this technology is going. Everyone is making their bets and is aware of the significant risks but they don't want to miss this wave of technology as they both confirm this is one of the most important innovations they seen in their careers.
What investors like Ed Sim are looking for when they evaluate a new start-up for investing:
Did they define their problem in detail? Do they truly understand this problem?
Who are the first five people joining the founder in solving this problem? Are they top talent and bring the needed skills? Are they motivated and determined to solve this problem as much as the founder is?
Is the founder able to tell a story and articulate that story well?
How is the start-up planning to use AI to solve this problem and provide value to its user base or customers?
Trevor Lee - Founder & CEO at Myko
Iman Oubou - Co-founder & COO Vocable
Peter Yared - Founder & CEO at InCountry
Peter Yared's main learning from his experience with multiple implementation across the world is the need to have a flexible, decoupled back-end architecture, so the models and other core elements like vector database solutions can be swamped at any time without major re-engineering. He mentioned OpenRouter as one way to handle this challenge.
The models and back-end architecture solutions are evolving so fast that within 3 to 6 months, you might have to swap LLMs or upgrade major architecture components of your solution. (As a top mobile senior engineer and friend of mine would say, "Don't bury the basement." Tobe Sullivan )
Trevor Lee, founder of Myko, learned the importance of meeting the customers where they are.
They all noticed a big gap between their customers' tech skills and familiarity with AI technologies and the state of the tech AI start-up industry today.
They all recognize the common challenge for their customer in adopting new tech solutions for typical enterprise users. Once users see the value and benefits, they never want to go back.
They all acknowledge the importance of taking the time to train their customers on their AI Agent solutions so they can minimize that tech capability gap.
They all notice a common request from all their customers using their AI Agent solutions, which is the importance of integrations with internal enterprise systems more than they have bandwidth to accommodate and the importance of prioritization. They also had many request their AI Agentic solutions to be integrated with common SaaS platforms. Integrations are a must.
My main observations and advice from these experienced founders are:
Ensure you are solving a real problem before introducing the AI Agent solution.
Do not commit to any LLM or sub-platform; be flexible as an AI Agent start-up.
Make sure your build team spends at least 30%—50% of its time evaluating the quality of your AI Agent solution. An evaluation framework or automated QA approach at scale is a must.
For AI Agents to expand into core business workflows or solutions, transparency on the training data and how the AI models are trained, as well as what the AI agents execute behind the scenes is very important to build trust with your customers.
Even with all the hype out there, they all agree the AI Agent race is just starting. We are very early in this space.
Most customers, both enterprises and small businesses, don't seem to be ready for the latest AI Agent tech solutions and are slow to adopt them. But once they do, they never want to go back.
Robotics or IoT embedded with an LLM/AI Agent seem to take the attention of some of these CEOs and founders.
No one knows what the future UI for AI Agents will be.
Some founders are wondering if their current platforms might be impacted or replaced by the new way of how AI Agents UI will evolve.
One thing is sure and they all agree on is that companies that are not at least experiencing in some form with this wave of AI technology will be left behind and might not be able to catch-up later on since this field is evolving at a speed they never seen before.
About me
Cristian, as the former Director of Delivery and Program Management, led the agile delivery and PMO across 60 data services and data science team members at FOX Corporation.
He has 12 years of experience leading software and data teams delivering large-scale complex projects. He is also a prior Agile Delivery and Product consultant, leading initiatives for large clients like Ford, Chrysler, and Mannheim, and he has a proven track record of delivering $20M+ projects for Fortune 500 companies.
Lately, Cristian has been specializing in AI automation and Generative AI solutions and is certified in:
- AI Product Manager
- Microsoft Azure AI Fundamentals
- Databricks Generative AI
- Latest Agile Delivery Frameworks for AI projects
If you are looking for an experienced technical delivery or product person to lead the execution of your next AI projects, reach out to me on LinkedIn.
Thank you to the Moderators for this great experience at the Miami AI Agent Summit:
Burhan Sebin - Moderator (Founder of Miami AI Hub & VP at Atlas Space)
Greg Isenberg - Moderator (CEO and Co-Founder of Late Checkout)
Ayal Stern - Moderator (Co-founder of Lab Miami)
If you are a Moderator or panel participant and want to add comments or changes to the above write-up, feel free to contact me directly with a LinkedIn message.
Yorumlar