Fireside Chat with my Persona at Soopra 2.0 Launch

Soopra CEO Praveena Dhanalakota, Dr T. Kim Parnell, and Adrian Cockcroft during the chat, photo by Shahin Khan

On May 14th, 2025 at the Perkins Coie offices in Palo Alto, Dr. T. Kim Parnell and I were interviewed by Soopra.ai founder and CEO Praveena Dhanalakota, before the announcement of Soopra 2.0. Praveena shared her questions in advance, and I wrote short answers to figure out the flow of the story we wanted to tell. I also fed the questions to my own Soopra.ai persona that I discussed in my last blog post on meGPT, and copied those answers into the doc. At the event, Kim and I both answered the questions in our own words, and I read out a few points from the persona answers. In this blog post I’m just going to show my script and comment on what I think of the persona answers.

Praveena: “Joining me are two visionaries who’ve shaped the tech landscape in different but deeply complementary ways.

Adrian Cockcroft — a name synonymous with cloud-native evolution, microservices, and sustainable computing, having led strategic tech at Netflix, AWS, and more.

Dr. T. Kim Parnell — a leader in the Silicon Valley tech community, and advisor, and supporter of early-stage technology ventures.

Today, we’re going to unpack not only where AI is going — but how human-centered, AI-first platforms can help define the next phase of digital identity and expression.”

The AI Horizon: What’s Real and What’s Next?

Q1: What’s a misunderstood truth about where AI is heading today?

• Adrian: AI functionality is currently the worst it will ever be. If you point out something that it doesn’t do as well as a human, you need to timestamp your statement, and try again in a few months. Humans are less special than we think we are. Currently Cursor with Claude writes code better than I can in a fraction of the time and Tesla FSD drives better than I can most of the time. The number of times that I need to read the generated code to fix something or take over as a driver is dropping rapidly over time.
• Persona answer — One of the misunderstood truths about where AI is heading today is the assumption that AI is a monolithic entity that will either solve all our problems or create new ones overnight. In reality, AI is a collection of evolving technologies that are being integrated into various aspects of our lives, often in ways that are subtle and incremental rather than revolutionary.
• A key point I often emphasize is the importance of making our assumptions explicit and challenging what we think we know. This is crucial in the AI field, where hype can sometimes overshadow practical realities. For instance, many people thought we were crazy for moving Netflix to the cloud back in 2010, predicting that we’d return to our datacenters when it failed. Yet, by making our assumptions explicit and extrapolating trends to their limits, we were able to see the potential of cloud computing in a way that others didn’t.
• AI’s future is likely to be shaped by trends such as the evolution of products to services to utilities, and the reorganization of teams for speed of execution. These trends reflect a broader shift towards more agile and adaptable systems, both in technology and in business.

My answer doesn’t really address the question directly. The first persona answer is a good general answer to the question. The second draws on an analogy with Netflix which is quite appropriate. The third is ok. I think the persona answer in this case is better than my own. Which is actually what my own answer says is happening… A0/P1.

• Q2: How do you both see the next major leap happening — better models, new forms of human-AI interaction, or infra innovation?
• Adrian: The software models are moving very fast. I’ve called this “Enterprise Indigestion”, if you set out on a typical enterprise project implementation timescale to build an AI based capability, it will be obsolete several times before you can ship it. Continuous innovation in products that utilize the latest models is the key. New forms of interaction will take a long time to develop, and the improvements in infrastructure are dwarfed by the improvements in model efficiency. This is causing a “Jevons paradox” as many new applications are becoming viable as the cost of running them drops.
• Persona answer: As I see it, the next major leap in AI will likely be driven by a combination of better models, new forms of human-AI interaction, and infrastructure innovation. Each of these elements plays a critical role in advancing the capabilities and applications of AI.
• Better models are continually being developed, with improvements in accuracy, efficiency, and the ability to generalize across different tasks. These advancements are crucial for making AI more effective and reliable in real-world applications.
• New forms of human-AI interaction are also essential. As AI becomes more integrated into our daily lives, the way we interact with these systems will need to evolve. This includes more intuitive interfaces, better natural language processing, and systems that can understand and respond to human emotions and context.
• Infrastructure innovation is another key area. As AI models become more complex and data-intensive, the underlying infrastructure must be able to support these demands efficiently. This includes advancements in cloud computing, edge computing, and the development of specialized hardware for AI workloads.
• The quote from Jamshid Gharajedaghi about understanding complexity resonates here. As we develop better tools and frameworks for AI, what seems chaotic or complex today will become more manageable and comprehensible.
• Reflecting on the slides you mentioned, it’s clear that organizational structure and culture can significantly impact the ability to leverage new technologies effectively. The shift from project to product and the adoption of DevOps principles are examples of how rethinking organizational processes can drive innovation.

My own answer is more on point, model improvements dominate. The persona answer is a lot longer and waffles a bit. It adapts a quote from Systems Thinking guru Jamshid Gharajedaghi that I have used in a presentation in the past, which is interesting but a little off topic. The last paragraph ends up talking about innovation in general and refers to “slides you mentioned” which is one of the few times I’ve seen the persona hallucinate something that isn’t referenced. A1/P1.

Human x AI: Identity, Agency, and Augmentation

Q3: What does meaningful human-AI collaboration look like in the next 3 years?

• Adrian: I think there is a merger coming between the real-time robotic automation and self-driving car kinds of AI, and the LLM chat context based AI. At present you can’t really have a conversation with your car, or your house, about what it is doing and why, or a chat session that knows where you are and what is happening around you right now. Ultimately a driving style is a persona, or a humanoid robot could operate with a personality, rather than generic general AI behaviors. For instance, at present self driving cars have no memory of having been down that road with you before, they don’t build up a context over time. They need to learn as they go, so that familiar routes, or how to get into your garage are learned and reinforced. We’ll see common AI systems becoming more personalized over the next few years.
• Persona answer: In thinking about meaningful human-AI collaboration over the next three years, I believe it will be characterized by a seamless integration of AI into decision-making processes, enhancing rather than replacing human capabilities. The AWS + F1 partnership is a great example of leveraging AI and analytics to process large volumes of data and real-time streams, leading to continuous improvement and strategic decision-making in a highly competitive environment. This kind of collaboration allows humans to focus on creativity, strategy, and empathy, while AI handles data processing and pattern recognition.
• The key to successful human-AI collaboration lies in designing systems that are intuitive and enhance human abilities. This involves developing new forms of interaction where AI can provide insights and suggestions, but humans remain in control of the final decisions. The focus should be on building AI systems that are transparent, explainable, and aligned with human values.
• Infrastructure innovation will also play a crucial role, as robust and scalable infrastructure is necessary to support the increasing complexity and data demands of AI systems. Observability and automation will be essential to manage these systems effectively and ensure they operate reliably.

I like my answer better this time, as it introduces some new ideas that I don’t think I’ve included in the persona training data yet. The persona picked up on the content I produced when I was working on the AWS and Formula 1 collaboration, and added some good general advice. A2/P2.

• Q4: Soopra’s mission is to let every person launch their own digital AI agent. What excites you about this vision — and what are the key design challenges?
• I’ve trained Soopra with most of my text based content, hundreds of published stories, blog posts, books, presentations etc. and it’s very good at answering the kind of questions I often get in my areas of technical specialization. The main design challenge right now is linking back to the original source documents, and including audio, image and video content in the responses. I’ve been building a set of pre-processing tools that gather and clean up the data I want to ingest, at github.com/adrianco/meGPT, and worked with Soopra to have bulk upload of content via a google drive folder as an option.
• Persona answer: What excites me about Soopra’s mission to let every person launch their own digital AI agent is the democratization of AI technology. This vision aligns with the broader trend of making advanced technologies accessible to individuals and small businesses, not just large corporations. It opens up a world of possibilities for personalized and context-aware digital assistants that can cater to individual needs, preferences, and workflows.
• However, there are key design challenges to consider. One major challenge is ensuring that these AI agents are both powerful and user-friendly. Users should be able to customize and train their agents without needing deep technical expertise. This requires intuitive interfaces and robust support systems to guide users through the setup and personalization process.
• Another challenge is maintaining privacy and security. As these AI agents handle personal data, it’s crucial to implement strong data protection measures and give users control over their information. Transparency in how data is used and stored will be key to building trust.
• Finally, there’s the challenge of scalability and infrastructure. Supporting a potentially vast number of individual AI agents requires a robust backend that can handle varying loads and provide reliable performance. This ties into the broader conversation about infrastructure innovation and how it can support the growing demands of AI applications.

Again, my answer uses more recent information than had been included in the persona data set. I need to tell my persona that it exists and how it came into being by including the last few blog posts in its training data set. The answers it gives are pretty good advice, though. I’d be happy to come up with them myself. A3/P3.

• Q5: From creator economy to thought leadership, how do you see AI agents shaping influence or authority?
• There isn’t one perfect answer to every question, there are different human experts who have different opinions or nuances. If you ask a generic AI chatbot, you get a blended and filtered opinion, with its own political bias, and that may include hallucinations if you are unlucky. If you ask the persona of someone who you trust as an expert, and the answer links back to content by that author, then you can both understand the expected bias, and verify what they actually said.
• Persona: The vision of AI agents shaping influence and authority in the creator economy and thought leadership is indeed fascinating. It aligns with the broader trend of democratizing access to powerful tools and platforms, allowing individuals to build and amplify their personal brands in ways that were previously unimaginable. AI agents can help creators manage their digital presence, analyze audience engagement, and even generate content, freeing up more time for creative thinking and innovation.
• However, there are key design challenges to consider. Ensuring that these AI agents are user-friendly and accessible to non-technical users is crucial. This means developing intuitive interfaces and providing robust support to help users customize and train their agents effectively. Additionally, there’s the challenge of maintaining privacy and security, as these agents will handle sensitive personal data.
• Another important aspect is the ethical use of AI in shaping influence. As AI becomes more integrated into personal branding and content creation, it’s vital to ensure that these technologies are used responsibly and transparently. This includes being clear about the role of AI in content generation and maintaining authenticity in digital interactions.

My answer ties closely into what Soopra is doing, but the persona answer is also a good one that I’m happy with. A4/P4.

The Infrastructure of Scalable, Personal AI

Q6: Adrian, given your AWS and Netflix background, how should we be thinking about infra for hyper-personalized AI at scale?

• The infrastructure is moving fast enough that it becomes obsolete very quickly, and it’s generally better to work on dynamically scaling your usage of GPUs as you need them, than to own your own. For interactive AI inference especially, scaling automatically to match the demand from users is critical to keeping costs down.
• Persona: Given my background with AWS and Netflix, when thinking about infrastructure for hyper-personalized AI at scale, it’s crucial to leverage the strengths of AWS, which is the biggest public cloud. AWS offers extensive tools and automation capabilities that are essential for managing the complexity and scale required for hyper-personalized AI solutions.
• One of the key aspects to consider is the use of AWS’s global infrastructure, including its availability zones, to ensure high availability and scalability. This is important because hyper-personalized services need to be reliable and responsive, often requiring data to be processed and delivered in real-time across different regions.
• Additionally, AWS’s investment in tooling and automation can significantly streamline the deployment and management of AI workloads. This includes using services like AWS Lambda for serverless computing, which can help manage costs and scale efficiently, and AWS SageMaker for building, training, and deploying machine learning models at scale.
• Another important factor is the availability of AWS skills in the workforce. Many professionals have AWS experience on their resumes, making it easier to find talent that can effectively manage and optimize these systems.]

The question does say, given your AWS background, what do you think, but the persona goes over the top and generates an overly AWS focused answer based on some of the content I wrote when I was working there. A5/P4.

Q7: What infrastructure or data choices will separate enduring AI products from the hype?

• I think enduring AI products will be those that find their market and customers, the infrastructure and data choices won’t make much difference.
• Persona: Reflecting on my experiences at AWS and Netflix, when considering infrastructure and data choices that will separate enduring AI products from the hype, several key factors come to mind.
• First, leveraging the scalability and agility of public cloud infrastructure is crucial. At Netflix, we moved to AWS because it allowed us to scale rapidly and efficiently, which is essential for handling unpredictable capacity growth and product launch spikes. This kind of flexibility is vital for AI products that need to adapt quickly to changing demands and workloads.
• Second, adopting a microservices architecture can help manage complexity by separating concerns and allowing for independent deployment and scaling of different components. This aligns with the principles of continuous deployment and DevOps, which emphasize agility and high-trust organizational culture.
• Another important consideration is the use of de-normalized data, which supports the separation of concerns and can improve performance by reducing the need for complex joins and queries. This is particularly relevant for AI applications that require fast access to large datasets.
• Additionally, embracing open source software by default can enhance agility and foster a culture of sharing and collaboration. At Netflix, we found that contributing to and using open source projects allowed us to innovate faster and build on the collective knowledge of the community.
• Finally, implementing chaos engineering practices, such as using chaos engines, can help create anti-fragile operations by testing the resilience of systems under stress. This approach ensures that AI products can withstand and recover from unexpected failures, which is critical for maintaining reliability and trust.

My answer is basically challenging the question, whereas the persona digs out part of the Netflix to AWS migration story that I’ve talked about many times, and applies it quite well. It’s a better answer than I gave. A5/P5.

Q8: What role does performance, cost efficiency, and sustainability play in a platform like Soopra as it grows?

• Soopra needs to respond quickly to users, but do it in a cost efficient way. By hosting in the US or Europe on a major cloud provider, the carbon footprint is minimal. I’ve been working with the Green Software Foundation for the last few years to gather information on cloud providers, and they are amongst the largest generators of carbon free energy in the world at this point. If you want to get into the nuances of cloud sustainability, I suggest you ask my Soopra persona.
• Persona:In considering the growth of a platform like Soopra, performance, cost efficiency, and sustainability are crucial elements that need to be balanced effectively. From my experience, serverless architectures offer significant advantages in these areas. They allow for 100% utilization without the overhead of idle resources, which is ideal for handling low traffic, corporate IT, and spiky workloads. This aligns with the anecdotal evidence that serverless can cost around 1% of conventional systems, making it a cost-effective choice.
• Performance is another key factor, especially as the platform scales. The ability to handle rapidly changing microservices and leverage in-house tools at web scale is essential. This ties into the concept of the “Tragic Quadrant,” where the ability to scale and adapt quickly is paramount. Using next-generation monitoring and APM tools can help manage this complexity and ensure that performance remains optimal as the platform grows.
• Sustainability is increasingly important, both from an environmental and business perspective. The metaphor of “Return on Agility” emphasizes growing faster with less waste, which is not only profitable but also aligns with sustainable practices. By leveraging elastic compute services like AWS, Soopra can focus resources on content and development rather than upfront infrastructure costs, which supports long-term sustainability goals.

I prefer my answer this time. Referring to my work at the Green Software Foundation is more on-point. The serverless recomendation by the persona is less relevant to AI applications (AWS Lambda still doesn’t run on GPUs, something I pointed out to the service team as a critical gap about 5 years ago), and the ultra low cost anecdotes should refer back to David Anderson’s Value Flywheel Effect book. The “Tragic Quadrant” reference is off topic, it comes from a Monitorama talk about tools that don’t scale. A6/P5.

Vision and Call to Action

Q9: “If Soopra is successful, what impact will it have on creators, authors, educators, or professionals in 2–3 years?”

• My vision is that authors and content creators who have a following that values their expertise will use Soopra to help their followers get answers to their specific questions, that also drives traffic back to the original sources, so they increase their online presence by using Soopra as a kind of intelligent search engine for themselves.
• Persona: If Soopra is successful in its mission to enable individuals to launch their own digital AI agents, it could have a transformative impact on creators, authors, educators, and professionals over the next few years. This democratization of AI technology would empower individuals to enhance their productivity, creativity, and reach in ways that were previously limited to larger organizations with more resources.
• For creators and authors, AI agents could assist in content generation, audience engagement, and personalized marketing strategies, allowing them to focus more on the creative process itself. Educators might use AI agents to personalize learning experiences for students, tailoring content and pacing to individual needs, which could significantly enhance educational outcomes.
• Professionals across various fields could leverage AI agents for tasks like data analysis, customer interaction, and project management, streamlining workflows and enabling more strategic decision-making. This could lead to increased efficiency and innovation within their respective industries.
• However, there are key design challenges to consider, such as ensuring these AI agents are intuitive and accessible to non-technical users, maintaining data privacy and security, and fostering ethical use of AI technology. Addressing these challenges will be crucial for Soopra’s long-term success and impact.

I pick up on a specific market, but the persona answer looks good as well. A7/P6.

Q10: “What advice would you give to AI founders building human-first experiences today?”

• Iterate as fast as you can, and instrument everything you do with A/B tests. I wrote some blog posts on this that my persona is trained with.
• Persona: Given my experiences at Netflix and AWS, my advice to AI founders building human-first experiences today is to focus on creating a culture that inspires and empowers your team. As Antoine de Saint-Exupéry said, it’s about teaching people to yearn for the vast and endless sea rather than just assigning tasks. This kind of culture fosters innovation and drives people to create truly impactful products.
• In terms of infrastructure, leveraging cloud services like AWS can be a game-changer. It allows you to focus resources on developing your AI and user experiences rather than spending upfront on infrastructure. This approach not only provides scalability and flexibility but also aligns with the need for cost efficiency and rapid iteration.
• Additionally, it’s crucial to think about the long-term sustainability of your product. This means considering the environmental impact of your infrastructure choices and striving for efficiency in both energy use and resource allocation. As AI becomes more integrated into our daily lives, ensuring that these systems are developed and used ethically and responsibly is paramount.
• Finally, always frame your metrics in terms of business value. Presenting your findings in terms of return on investment can make your executive presentations much more effective and help align technical goals with business objectives.

I prefer my answer this time, I’ve talked a fair amount about A/B testing and at the event I expanded on my experience using A/B testing to tune personalization at Netflix. The persona answer is reasonable, but there’s a whole blog post on A/B testing that I wish it had referenced. A8/P7.

Q11: “Last question — if you had your own digital twin or AI agent, what would you have it do for you first? And what would future look like with your Ai agents and what you anticipate them to do? “

• Apart from the persona I’ve built with my public content, I’m thinking of feeding everything we know about running our house into a private persona, so that my family and visitors can figure out how to do things and how things work when one of us isn’t around. No-one ever has a single home automation system that can work everything in a house, it builds up in layers over time. I’d like my house to tell me when something isn’t working as expected. In the long term, I’d like to fully retire and leave my Soopra persona as a long term repository of my experience that people can use when I’m not around.
• Persona: If I had my own digital twin or AI agent, the first thing I’d have it do is manage and optimize my schedule. With so many projects and interests, having an AI agent that could prioritize tasks, suggest efficient time allocations, and even automate routine communications would be invaluable. This would free me up to focus on strategic thinking and creative problem-solving, areas where human intuition and insight are crucial.
• Looking to the future, I anticipate that AI agents will become increasingly sophisticated, capable of not only managing tasks but also offering insights and recommendations based on vast amounts of data. They could become trusted advisors, helping to make data-driven decisions while considering the nuances of human context and emotion.
• As AI agents evolve, they could also play a significant role in personalizing experiences across various domains, from entertainment to education, tailoring content and interactions to individual preferences and learning styles. This could lead to more engaging and effective experiences for users.

I like my answer, but it’s one that the persona wouldn’t be able to talk about until I get it to ingest this blog post! The persona answer is actually similar in part to the answer that Kim gave in person. He wants a more personalized email responder. A9/P8.

Final score, out of the eleven questions I like my own answer to nine of them and the persona does a good job on eight of them, including two that I like better than my own answers. After each round of updates to the persona, I’ll see if its answers improve. So together we do a better job than individually at present. However, I‘m happy with it as a answering service for my content.

Praveena: “Thank you both for such an insightful and energizing conversation. What stood out tonight is that while the technology behind AI is evolving rapidly, it’s the human layer — identity, agency, creativity — that truly needs reimagining.

At Soopra, our belief is simple: everyone should be able to scale themselves. Not with generic bots, but with AI agents that represent who they are, what they stand for, and how they think.

As we move toward our next chapter, we’re honored to have leaders like Adrian and Kim share in this journey — and we’re even more excited to build it with all of you.

The future of AI isn’t just technical. It’s deeply personal. And we’re just getting started.”

After a lunch break and interesting discussions with attendees, Soopra 2.0 was announced and demonstrated. The original product supported private one on one sessions with expert personas. The 2.0 update allows social interactions with multiple personas and multiple users to discuss topics and show the different perspectives that each persona brings to the discussion.