Quantify 360 Solutions

OpenAI just launched their official agents framework - if you’re an AI agent developer or builder, here’s everything you need to know (coming from a founder building their own agents platform): => What is it? OpenAI “Agents SDK” is an open source agents framework with a core abstraction of agents + handoffs. For example: user sends a message “Hola, como estas?”, which gets sent to a “triage agent”, which then routes the request to the “Spanish agent” instead of the “English agent”. => Is OpenAI “Agents SDK” open source and model-agnostic? “Agents SDK” is open source and can be configured to work with third parties that support the ChatCompletions API. However - OpenAI’s new agent tracing is not open source. To use tracing, you have to log into OpenAI’s website and view their tracing dashboard. Remind anyone of LangChain (OSS) + LangSmith (closed source / SaaS product)? Additionally, OpenAI recommends users of the “Agents SDK” use their new Response API (which stores message histories, similar to the Assistants API). Currently, OpenAI is the only provider that supports this API. => What does this mean for other agent frameworks? Direct competition from the model providers. As the model layer commoditizes, OpenAI is starting to make plays in the framework layer. => What’s the difference between OpenAI’s agents framework and other open source agents frameworks? Forced “Handoff” abstraction: Like all abstractions, useful for some use-cases, and not a great fit for others. State is a second-class citizen: Similar to most agents frameworks, memory and state are not first party concepts - agents in OpenAI's Agents SDK do not have built-in support for long-term memory or context management. It’s expected that you run your agent script once, then “throw it away” - similar to a workflow. => What’s next for OpenAI? It makes complete sense for OpenAI to be moving up the agents stack as the model layer commoditizes - "Agents SDK" is just an initial step. My take: OpenAI's end-game (on the developer side) is a fully hosted stateful agents service, similar to their Custom GPTs or Assistants API. I expect to see a revised version of the Assistants API launch soon, likely rebranded as an “Agents API” (a paid hosted service that can integrate with the open source “Agents SDK”). => Why will / won’t OpenAI win? The problem for OpenAI is that developers do not want model or data lock-in in their agents stack. Developers want to be able to move their agent state (messages, tools, user data, data sources, memories) across model providers with zero friction. This is why provider-agnostic frameworks will win out over first party options made by OpenAI (or other frontier labs). Imagine if you built a verticalized agent startup on the OpenAI Assistants API in 2024. Your entire app is locked into OpenAI with no escape - no Sonnet, no DeepSeek V3/R1. The rise of stateful agents (e.g. agents with long-term memory) makes model lock-in even more devastating.

Here are the first few minutes of President Trump's remarks at today's White House Crypto Summit. What a fascinating conversation unfolding!

Yesterday, I had the privilege of attending an insightful event hosted by The GenAI Collective focused on rethinking the AI landscape at Microsoft Reactor. It was a deep dive into the current state and future potential of artificial intelligence, and I left feeling energized and inspired at all of the new ways people are thinking about AI. Here’s a quick recap of what I found meaningful: Matt F. led a fascinating discussion facilitated by Tian Lan. Matt emphasized that "the purpose of AI is not to overtake but to augment." He gave a compelling example of creating a personal AI agent to assist with his child's math homework. He also explored how AI agents are already impacting search results and the potential disruption in sectors like data analytics, DevOps, QA, and paralegal. I had the chance to ask about the state of each of AI landscape's current layers (infrastructure, LLM, consumer tech), with Matt highlighting consumer tech as the least saturated and most ripe for new solutions. It seems he agrees with the growing need for impactful AI consulting! ?? Yujian from OSS4AI provided a comprehensive overview of the open-source AI world, highlighting companies like Comet, TensorWave, Ollama, Lamini, Together AI, Fireworks AI, and LlamaIndex to name a few. It was also nice to see several Seattle-based companies on the list like BAML, Pixeltable, and CopilotKit??. ?? The keynote speaker, Trent Gillespie, offered a powerful perspective on innovation, defining it as "sustainable value from something new." He contrasted true innovation with practices that harm the environment and emphasized AI's role in accelerating the innovation process. ?? Several startups (Clarify, GenSX, Certifi AI, Breadcrumb.ai, Graphlit, by Unstruk Data, codeValet, and Viata) pitched briefly. Too much awesome stuff to outline here; check each of them out yourself! ?? It's clear that we're not just witnessing technological advancements, but the dawn of a new era, where AI empowers us to create more sustainable and impactful solutions. I'm eager to explore how these emerging technologies can be leveraged in helping existing businesses adopt AI and revolutionize their workflows. It's without a doubt an exciting time to be involved in tech! #ai #seattletech #meetup

The autonomy of AI agents makes it hard to pinpoint responsibility when mistakes occur — here’s what we recommend in the latest edition of Do More With AI. ???

Web3 was built on trust, transparency, and decentralization-not hype and exploitation. It’s time to build with purpose, create real value, and shape the future with integrity.

I attended #LlamaPALOOZA last night, one of the most fun and inspiring tech events of the year so far. The Labour Temple - Seattle was an incredible venue. And the list of speakers was packed with some heavy hitters. The speakers: - Sarah Chieng, from Cerebras Systems. Cerebras is redefining the limits of AI compute with their Wafer-Scale Engine. Their focus on accelerating the training of massive models is crucial for future progress. Look out NVIDIA! - Justin Lee, from Meta. Happy birthday to Llama! Llama models have been downloaded over 600 million times on Hugging Face alone. Open sourced AI is alive and well! - Darko Mesaro?, from Amazon Web Services (AWS). Amazon Bedrock is making generative AI accessible and scalable for businesses via a fully managed service for building with foundation models. - Patrick Devine, from Ollama. Ollama democratizes access to LLMs by making it easy to run them locally, which can be a game-changer for developers and researchers. Super inspiring to see such an incredible turnout. The guest list had over 700 people wanting to attend, and many were turned away due to capacity. Grateful to spend time networking and meeting brilliant people in the Seattle tech world! #ai #seattletech #meetup