A daily digest of practical AI agent workflows for non-developers. We isolate actionable no-code guides from viral hype. Scored against human-defined standards.
Daily Summary
3 curated | 3 evaluatedThe no-code AI landscape is evolving rapidly with new tools that make automation more accessible: , a vision-language model that understands editable design structures, while discussions emerged about and building agents through . These developments highlight how AI-powered automation is becoming both more specialized and more accessible to non-technical users.
🎨 Canva just launched its own AI design model that can edit an entire presentation, write code to modify layouts, and visually check its own work. Meet Canva Brush, a new vision-language model developed by Canva Research. Unlike ordinary image generators, Brush understands the structure of an editable design, including its layout, text, layers, colors, and visual hierarchy. And according to Canva, it's up to 69% faster and 46% cheaper per editing task than its previous AI design agent. 🖌️ What can Brush actually do? ▪️ Precise edits: Change a single line of text, adjust an element, or modify a particular section without rebuilding the entire design. ▪️ Presentation-wide changes: Restyle dozens of slides through one instruction while maintaining a consistent visual direction. ▪️ Format adaptation: Turn presentations into social media graphics, adjusting layouts rather than simply cropping. ▪️ Visual restyling: Change colors, backgrounds, imagery, and the overall aesthetic. ▪️ Fully editable results: Every element remains accessible in Canva's regular editor. 🧠 The interesting part: Brush can inspect and correct its own designs. Instead of generating one result and stopping, it uses an agentic execution loop: Understand the request and existing design. Write code to modify the relevant elements. Execute those changes. Visually inspect the updated composition. Refine the result when necessary. This allows Brush to consider not only what needs to change, but also what should remain untouched. ⚙️ How was it trained? Canva didn't build the foundation model from scratch. Brush starts from an unnamed open-weight model, further trained on design-specific tasks. Its training combines: ▪️ Real design and editing workflows ▪️ Anonymized data from users who opted into AI improvement ▪️ Large-scale reinforcement learning ▪️ Hundreds of thousands of simulated interactions ▪️ Feedback standards established by professional designers The reinforcement-learning reward system evaluates visual quality, adherence to design principles, and successful task completion. 📊 Performance In Canva's internal evaluations: ▪️ Editing speed: Up to 69% faster ▪️ Cost per editing task: 46% lower ▪️ Quality: Matched or exceeded the previous agent on most evaluated tasks The largest improvements appeared in restyling and adapting designs to different formats. These are Canva's own measurements. The company hasn't published independent benchmark results, model size, or the complete evaluation methodology. 🚀 Availability Brush is available now for supported editing tasks in Canva AI for Pro subscribers. To use it: ▪️ Open an existing design or attach one to Canva AI. ▪️ Select Brush from the model dropdown. ▪️ Describe the edits you want. ▪️ Continue refining through chat or switch to manual editing. ⚠️ Still an early release Canva is already working on several missing capabilities: ▪️ Brand Kit integration ▪️ Using attachments as references ▪️ Better handling of image-heavy layouts Broader availability is planned, but no date has been announced. 💡 Why this matters Most generative AI tools are good at creating a first draft but struggle with precise, repeated revisions. Brush focuses on the part of design work that often takes the longest: editing, refining, and maintaining consistency across an entire project. And unlike a flattened AI-generated image, the result remains a real, editable Canva document. The shift is from AI that generates designs to AI that works inside the design editor alongside you. 🔗 Official announcement: https://t.co/qP9jtMqspy 🔗 Try Canva AI: https://t.co/g7WKJuH49x #Canva #GenerativeAI #AIDesign #AITools
What it's really like to go from a 100+ person team to 2 in the AI era.... As you likely know, I stepped down as CEO of Jones Road earlier this year for several reasons. It wasn't one of the main ones, but one reason was that I am super AI pilled and I wanted to make sure I didn't miss this opportunity. I see this as the most important technology in my lifetime, and frankly one I really enjoy playing with. There wasn't enough time to do so as CEO. Here is where I see the state of AI in different company sizes: - The larger the company, the worse they are at AI generally. Enterprise cos are only budgeting $20 a month per employee; so it's hard to get stuff done. They also have much stricter security protocols so adopting or integrating new tools is hard. I'd say these companies are dabbling and much of it is performative. Productivity gains are small here for the most part. - Midsize companies are still slower to adapt. JRB was in this range, but at the lower end of it. These companies have existing tools, workflows, and enough resources to generally not need to lean on AU too much. To them, AI adoption looks like building dashboard, making data analysis much faster, and accessible to more people. The more ambitious of these companies are using it to increase creative velocity and to offload manual tasks. But it can be tricky due to existing tools and workflows. So it's a lot more analysis than creation. Not fully "agentic". - Then you have newer, AI first companies that know nothing else. They don't have people, so there's so existing debt to work around, and enough ignorance and naitive to think things can be done. Winks falls into this camp. Before hiring anyone, we prove to ourselves AI can't do it. Even if AI can't do it fully, it can likely really help. One person in these companies can get A TON done. I am one of them. I went from having 50+ people under me to 1. So if I want to get a lot done, I have no choice but to lean in. Call it the era of the Super Contributor. A few ways I've been able to be a former CEO managing an entire marketing function with AI: - Context and access is everything. The models and agents are only as good as what you give them access to. We use HQ for this. I couldn't imagine a world without it with how many agents we run. - We make use of many different agents and models. I always find it super cringe when people show their agent org charts but it's true. I have a Grokbot Chief of Staff that oversees all my other agents. It runs all project management in Notion. I literally never have to go into Notion and can do a lot of the work with my main COS Grokbot, who can sub work out to other Grokbot agents as well as other agents such as Claude Code, Codex, Runneth, Muse, etc. This sounds like overkill and it might be, but they each have a different strengh. This is only possible because all the agents have the shared context and have knowledge of everything like employees in a company would. - I've learned even AI pilled people need reminders to push what their agents are capable of. For everyone who says AI is not helpful, there are people doing it. They just have trained it more and pushed the boundaries. I uploaded ads for a few weeks manually until I found out Runneth can do it very easily. And it took some time to feel comfortable with it. Video editing is also solved, but don't expect it to be one shotted. - In addition to agent-to-agent comms, I believe humans and agents are going to collaborate much more in the future. For example, Notion is a solid platform that allows for a good UX but it also created markdown for agents to use. - Prompt engineering is dead. I see too many people still focusing on it. Instead, I like to use long-form Wispr flow yapping, with full context, multi-agent review workflows, and verification. Plus being in the loop as little as possible. - Your skills as experts in your field are the most important thing. There are people that can design amazing websites in AI because they are amazing designers without AI. Same with images and videos. It's why I am going back to focusing on learning copywriting and creative strategy almost exclusively. - We are still extremely early. If you feel behind, it's absolutely not too late. I truly believe we're in the first inning. Some will be left behind, but I think this is the best time in history for skilled and ambitious people who are willing to take this on. Your earning potential and career prospects can get taken to new heights if you use AI to leverage your unique expertise and skills. If you'd like to be one of those people, or if half of what I said above was foreign and you'd like to learn more, I have good news. I have teamed up with my friend @jacob_posel and the good folks at @HQForWork to help teach The Super Contributor Method, which is a live 4 week Cohort Based course teaching you how to level up your AI skills and become a super contributor. Spots are going fast and are limited. Sign up here: https://t.co/KK6VYEZKnP
Just found a pretty cool open-source project called Langflow. It lets you build AI agents and RAG workflows by simply dragging and connecting components. No need to write everything from scratch. Models, prompts, vector databases, and tools are all modular. Want to switch models or update a workflow? Just swap a component. The best part? Your workflows can become APIs and MCP tools automatically. That means Claude Code, Cursor, and other AI tools can directly call the agents you've built. For example, you could build an internal knowledge base with Langflow, connect it to Claude Code, and let it search your company docs while coding. There's also a new AI assistant that helps you build and modify workflows using natural language. I think this is where AI development is heading: Don't just build agents that complete tasks. Build reusable tools that other agents can call. Definitely worth checking out if you're building AI products or multi-agent systems. GitHub: https://t.co/FwYs7KzYGU