Tony Ojeda

Most people still use AI the same way: open a blank chat, paste context, explain the task, correct the output, and then repeat the whole process again on the next problem. I got tired of paying that setup cost over and over, so I built something else: a personal AI operating system with 25 specialized agents, 117 reusable skills, and memory that carries forward the standards and preferences that matter in my work. It now handles code review, research, writing, planning, and analysis in a way that feels much closer to working with a team of specialists than with one generic assistant. The interesting part is not the number of agents. It’s the design principle. AI gets more useful when you stop treating it like a smarter chat window and start treating it like a structured system with specialization, memory, and clear judgment boundaries. I wrote about what I built, why I built it this way, and what changed once the agents were real. tojeda.com/blog/i-built-a…

Most people still use AI the same way: open a blank chat, provide context, explain the task, correct the output, and then repeat the whole process again on the next problem. I got tired of paying that setup cost over and over, so I built something else: a personal AI operating system with 25 specialized agents, 117 reusable skills, and memory that carries forward the standards and preferences that matter in my work. It now handles code review, research, writing, planning, and analysis in a way that feels much closer to working with a team of specialists than with one generic assistant. The interesting part is not the number of agents. It’s the design principle. AI gets more useful when you stop treating it like a smarter chat window and start treating it like a structured system with specialization, memory, and clear judgment boundaries. I wrote about what I built, why I built it this way, and what changed once the agents were real. tojeda.com/blog/i-built-a…

A lot of enterprise AI teams are still trying to fix reliability problems with relevance work. They improve prompts, rework chunking, add examples, and expand the knowledge base. Sometimes that helps. But sometimes the model already had the relevant information and still produced the wrong outcome because the real failure was not retrieval. It was system design. The workflow let the model answer a question it never should have been trusted to answer on its own. That distinction matters more than many teams realize. Relevance asks whether the model had and used the right information. Reliability asks whether the system produces the right outcome at scale, including the cases where the model should stop, defer, or hand off. Those are different problems, and they need different fixes. I wrote about that distinction, why teams keep working on the wrong layer, and a simple test for diagnosing which problem they actually have. tojeda.com/blog/the-diffe…

Wanted to share that my third book, Collaborative AI, was just published. 🚀 It's about designing AI systems that actually work in production — the frameworks and architecture patterns I've been thinking through for the past few years. The book covers a Five Pillar framework (AI, Logic, Data, Evaluation, and Human judgment), four architecture patterns for different problem types, and the data strategy and governance questions that actually determine whether these systems work long-term. Available in Paperback, Kindle, PDF, and ePub. tojeda.com/collaborative-…

In my spare time, I've been developing a collaborative AI workflow for creating professional-grade, researched intelligence reports on just about any topic. Here's how it works... Research Phase: 🔍 My research agent conducts 20-25 different web searches (250+ sources) covering historical context, current events, and specialized areas 📊 Synthesizes findings into comprehensive research report with executive summaries, timelines, and market analysis using the latest data 💬 Obtains feedback from the user (me) on inaccuracies, gaps in coverage, and personal experience with the topic (if applicable) Analysis Phase: 🧠 My analyst agent transforms research into professional intelligence analysis and scenario modeling 🎯 Creates future scenarios with probability assessments and uses rating scales for all metrics 📈 Generates intuitive diagrams and visualizations based on its analysis Writing Phase: ✍️ My reporter agent creates 5,000-8,000 word strategic intelligence reports based on the research findings and analysis 🎨 Integrates all visualizations into flowing narrative with sophisticated analytical structure Editing Phase: ✅ My editor agent performs fact-checking, source verification, and flow optimization 🎭 Solicits user feedback on the report and incorporates it into its edits ⭐ Ensures final reports meet professional standards with actionable insights I've been pretty impressed with the depth and quality of these reports - they're great for quickly getting up to speed on complex topics or preparing for strategic discussions. Here's the link to one I created on #GenerativeAI: tojeda.com/the-algorithm/… I plan to generate these for more granular topics within AI as well as for other topics of interest that are not related to AI. What topics would you be interested in seeing a deep-dive like this for? Organizations mentioned in the report include @OpenAI, @AnthropicAI, @GoogleAI, @Microsoft, @xai , @nvidia, @GroqInc, @Meta, @huggingface, and more.

🎯 Finally built this prioritization app I've been thinking about for years! After seeing a similar concept online many years ago, I've been wanting to create my own prioritization tool that goes beyond basic ranking. This past weekend, I finally carved out the time to make it happen. The Prioritizer helps you make tough decisions through smart pairwise comparisons. It shows you two items at a time so you can focus and decide which is more important to you. Behind the scenes, it tracks the choices and ranks the items intelligently. Then it uses AI to provide personalized insights about your decision-making patterns and blind spots. Some of the things you can prioritize with this are: - Home improvement projects - Career goals - Life decisions - Business initiatives Building this reminded me why I love coding with the tools we have available today - taking an idea that's been percolating for years and bringing it to life in a weekend feels incredible. What decisions are you struggling to prioritize right now?

I built an AI system that mirrors how great teams solve hard problems. Instead of using a single AI model, it combines three specialized agents (a Domain Expert, a Creative Problem Solver, and a Critical Analyst) that challenge and build on each other's ideas. The strategies it produces combine high-level thinking with practical, actionable tasks surprisingly well! The demo video in the post shows it coming up with a plan to optimize a retailer's supply chain. Would love to hear thoughts and feedback! Check it out here: tojeda.com/the-algorithm/…

Automation is evolving. AI Agents bring flexibility, Software Agents bring precision, and together, they’re solving challenges neither could handle alone. tojeda.com/the-algorithm/…

soo the Perplexity API is good and cheap?!

Getting all of my thoughts down and organized is always the hardest part when writing something. At least for me. I'm playing with an interface that you can speak to. It will jot down index cards as you're rambling and organize them into main topics.

Google's Veo 2 and Imagen 3 are pushing AI creativity to new heights! From stunning 4K videos to photorealistic images, these tools are revolutionizing content creation across industries. Learn why this matters and what it means for the future of AI. tojeda.com/the-algorithm/…

Gemini 2.0 is here! Google’s latest AI breakthrough combines multimodal capabilities, advanced reasoning, and autonomous task execution. Why does this matter? It’s a leap into the future of agentic AI. Dive in to explore the possibilities: tojeda.com/the-algorithm/…

OpenAI just wrapped up 12 days of remarkable announcements: advanced reasoning models, text-to-video generation, real-time AI collaboration tools, and more. Here’s a recap of how they’re shaping the future of AI. tojeda.com/the-algorithm/…

Imagine having the brightest minds in history working together to solve a difficult problem, at mind-blowing speed.💡🚀 Last week, I shared an initial prototype of my “debating agents” app. This week, I'm excited to introduce an enhanced version that not only allows the agents to debate 🤼‍♂️ but also to collaborate 🤝 with each other. With this new update, users can choose between two modes: Collaborate or Debate. Whether you want to see agents build on each other's ideas to find the best solution or engage in a spirited debate to reach a logical compromise, the app does both. Other Key Enhancements: 💬 Streamed Responses: Agent responses stream in real-time, making the conversation more natural and engaging. 📋 Updated Rules: More comprehensive rules for collaboration and debate modes to effectively guide the conversation. 🔍 Enhanced Final Output: More comprehensive solution or well-reasoned compromise based on the selected mode. Special thanks to the tech stack that made this possible: - @AIatMeta for providing the powerful and flexible conversational Llama model. - @Groq for providing the infrastructure to run inference at lightning speed. - @Streamlit for providing the platform to build the app. #AI #GenerativeAI #GenAI #ConversationalAI #Collaboration #Debate #Innovation #Tech

If you haven't already checked out @Google's NotebookLM, you should. 📝 You can upload documents and get different types of summaries (FAQ, study guide, etc.). 💬 You can chat with it and get answers based on the content in the documents, with references that show you where each part of the answer came from. 🎧 But the most impressive feature is that it can create an AI-generated two-person podcast where they discuss the topics covered in the documents. 🤯 Have you used it? If so, what do you think? If not, give it a try and let me know your thoughts! blog.google/technology/ai/… #GenerativeAI #AI

g1 is experimental and being open sourced to help inspire the open source community to develop new strategies to produce o1-like reasoning. This experiment helps show the power of prompting reasoning in visualized steps. hubs.la/Q02QbTHg0

I used @GroqInc with @AIatMeta's Llama 3.1 to create a fun little @streamlit app where I can have any two people debate any topic, have as deep a discussion about it as I'd like to see, and then come up with a compromise at the end. 🤝 I like that it lets you see both sides of an issue and how a nuanced compromise can be arrived at that respects both sides of an argument. It's also entertaining to plug in different characters as the debating agents and see the language they use! #GenerativeAI #AI

Excited to share a new video showcasing the latest enhancements to our AI-powered data analysis assistant app, along with an insightful analysis of @Apple's financial statements! 🍎🚀 Recent Enhancements: 📈 Time Series Analysis: The app now excels at analyzing time series data, providing deeper insights into trends over time. 🤖 Self-Learning from Errors: When the generated code encounters errors, the app feeds the errors back into itself to learn and improve. 🔮 Forecasting Capabilities: Using @Meta's Prophet library, the app can now generate forecasts for future performance. Key Insights from Apple's Financial Data: - Revenue Growth: Apple's revenue peaked at $394.33 billion in 2022, with a forecast to reach $547.58 billion by 2028, showcasing consistent sales growth. - Profit Margins: Gross profit margin improved from ~37.5% in 2014 to around 45.5% in 2024. Net profit margins also increased, indicating effective cost management and profitability. - Operating Expenses: Significant rise in operating expenses, particularly in research and development, highlighting Apple's investment in innovation and market expansion. - Earnings Per Share: Through buyback programs, the number of outstanding shares has been reduced, positively impacting earnings per share, which increased from $1.44 in 2014 to $6.45 in 2024. - Financial Health: Strong revenue growth, efficient expense management, strategic investments, and a commitment to returning value to shareholders through consistent dividend payments and share buybacks. Check out the video to see how quickly and efficiently the app performed this analysis. 🎥👇 Stay tuned for more updates, and let me know if you're interested in a live demo or customized version of this for your company! 📊✨ #AI #ArtificialIntelligence #GenerativeAI #DataScience #DataAnalysis #Analytics #MachineLearning #TechInnovation #Apple #Meta #FinancialAnalysis #TimeSeries #Forecasting #Innovation