Ram Venkatesh

At Sema4 we have found the agent UX and the agent supervisor UX to both be valuable for humans and agents to collaborate. Evaluating the work for alignment with objectives, or, its corollary, divergence, is greenfield from a UX standpoint but its importance is clear from our first set of production deployments

AI Agent UX is one of the most interesting design questions today. It’s clear that for now you need a high degree of visibility into the work the agent is doing, and need to have the ability to interact with it at a granular level. There are lots of use cases for AI Agents where the longer the task, the more the agent can go off the wrong path, and where subtle intervention is necessary for it to head in the right direction. Most agentic UX right now is built for a world of these brief bursts of work that a user can fundamentally review and control in-line. However, as models, tool use, and context continue to improve, it’s clear that AI Agents will be able to handle longer running tasks with less frequent interaction. Over time, the gaps between the points with lengthen. The question is how much of the UX changes at that point? Interestingly, the UX of passing work between people doesn’t change all that much depending on the level of expertise on either side — for instance, whether you’re working with an intern or the deepest expert in a field. What changes, instead, is usually the amount of context that is sent in the interaction and the size of chunks of work you get back. So we could certainly imagine a future where today’s paradigm of interacting with AI Agents largely stays the same, and the UX you build today scales for continued model improvements over time. But what happens when you have threads going with dozens or hundreds of AI Agents doing work for you in the background? What happens when the unit of work becomes much larger than it is today? What happens when AI Agents aren’t just triggered by an end user kicking off a task, but instead an embedded into a workflow happening behind the scenes? Lots of fun challenges in designing the future right now that we’re only starting to grapple with.

@gokulr The crux of your argument is about the change implied by the agentic model, MCP just makes it 10x easier. If the agentic model gains traction, there is a new layer between the business user and the apps regardless of what the app vendors want. Cue agent-washing apps now.

THE INEVITABLE MCP BATTLE If you work in AI, you've probably heard of MCP (Model Context Protocol), an extension of tool calling that allows agents to interact with applications in a standardized way. Social has been full of glowing reviews of MCP. Standards are awesome, so why wouldn't everyone embrace it? Two simple, logical reasons that MCP will not be welcomed by the largest and most ambitious applications: (a) Customer relationships and (b) Ads (a) Any B2B or B2C app that's built deep direct customer relationships over years and decades is going to view MCP as a disintermediatory force, one that decisively makes the AI agent the owner of the relationship. The analogy to search is uncanny: MCP allows AI agents to index apps, just like Google indexed websites. And we know who ended up winning that one. (b) Any app that monetizes their audience through ads will be even more directly impacted negatively by MCP. Every user who uses an agent to access the app through MCP, instead of logging into the app directly, results in lost ads engagement and ad revenue for the app. Ads is the highest margin revenue stream for many commerce apps, one they will strongly resist any attempt to diminish this business line. Apps who've always enabled API access, should likely not have any problems with MCP. However, UX-centered apps that have never opened up API access, owe it to themselves to consider the factors above before jumping headlong into MCP. I'm curious if any of the AI agents who are touting MCP, build an MCP server that allows other AI agents to access their data. ;) This will be an interesting tug of war.

@langchain Thank you @langchain Appreciate the shout-out and looking forward to the session tomorrow!

🤖OpenGPTs Hacking Hours Tomorrow, Friday March 15th, 4:30-7:30pm PST Join us and our friends at Sema4.ai for hacking hours dedicated on OpenGPTs! We'll show some cool new features, and discuss the future of the project. We are excited to get your feedback, and plan the future of open-source GPTs together. It's also a bit of special session, as this will be the first "public" appearance of the Sema4.ai crew! Their co-founder Ram, ex-CTO of Cloudera, will talk a few minutes why they started a new company, why open-source is important and the work they do with LangChain. rsvp: lu.ma/pk10ix8w

@JamesMaguire Thank you @JamesMaguire this was fun!

Thanks all! This has been another excellent #eWEEKchat. Serious insight today – great to see this monthly gathering. Stay tuned for next month’s chat!

A8. Baby steps. Don’t start with the ‘family jewels’ as your first deployment. Your people need to build up an understanding of applicability, alignment and steering LLMs which will take time. #eWEEKchat.

A6. Context matters. A lot. Use patterns like Retrieval Augmented Generation to safely inject enterprise context into your LLMs. #eWEEKchat.

A7. Think about a future government audit of your AI-driven decisions; would they hold up for privacy and fairness? #eWEEKchat

@PhilipAWCooper @JamesMaguire ^This. GenAI is a great example where controlling your data can create value that you were not aware of six months ago

@JamesMaguire A4. Isolate and protect your data. #eweekchat

Q4. How do you recommend companies address these generative AI challenges? #eWEEKchat

A4. Work on the use cases outside in, build up competency, think long term. Having a robust AI strategy for the long term is going to be more valuable than any short term hype cycle considerations. #eWEEKchat

A3. All of the above. Data & analytics have two decades of what good governance looks like, the entire generative AI stack needs the same treatment, from collecting and managing the data to audit and traceability of outcomes. #eWEEKchat

A2: Many of our existing customers with mature AI/ML programs are accelerating their efforts to include genAI. Their investments in AI are validated and now suddenly “mainstream” #eWEEKchat

A1. It has already disrupted as mindshare, now whether it lives up to its hype is up to us. No pressure. #eWeekchat

Hi everyone, Ram Venkatesh CTO @ Cloudera here. On with @JamesMaguire on Generative AI with #eWEEKchat

If you are not multicloud, you are all-in, choose wisely #eweekchat crowdchat.net/s/16a8b

A8: Portability will get easier. Data egress costs will come down (come on lets be bold, we have the Easy Button). Multicloud will get simpler as a consequence. #eweekchat crowdchat.net/s/16a84

Autonomy is like entropy, there is only one direction in reality, central teams need to accept this and plan for it. More governance, less mandates #eweekchat crowdchat.net/s/46a7x

A5 Establish the right metrics so you can tell if multicloud is the right option for your company, and if your implementation is actually yielding the results you are after. #eweekchat crowdchat.net/s/96a72

A5 @bernardgolden great point. Don't use least common denominator thinking to stay "cloud neutral", its about "cloud being "cloud leveraged" to help your business get the best outcome #eweekchat crowdchat.net/s/76a6t

If your use cases warrant multicloud, embrace it. Optimize for the business outcome rather than platform feature / function type tradeoffs. #eweekchat crowdchat.net/s/86a6l

@JamesMaguire Open source has its challenges too, but in this context, its powerful, especially compared to the alternative #eweekchat crowdchat.net/s/46a6c

Q3. What are the most pressing – and most common – multicloud challenges today? #eweekchat crowdchat.net/s/06a5g