Ryan Fox-Tyler

I’ve teamed up with @jamesm on various projects at Tailwind, Clerk, and other contracted gigs. We’ve chatted a lot about wanting to teach, share, and give back to the design community. Shape FM is our new podcast to influence your design and design engineering careers.

What if building AI agents felt as natural as writing code? 🤔 Discover how Hypermode’s “stay-in-language” approach is redefining agentic development—no more context switching, just pure productivity. Read the blog 👉 hypermode.com/blog/stay-in-l… #AI #AgentInfrastructure #AIAgents #NaturalLanguage

Next up - Briefing in @hypermodeinc @ryanfoxtyler and team - it is all about managing and orchestrating agents. With @mikeni @BarbaraCos25450 #NextGenApps #Briefing

@amsalem_lior @hypermodeinc We're working on a new serverless mode for Dgraph that will allow us to provide a richer free tier! waitlist: hypermode.typeform.com/serverless

@hypermodeinc Hey, any chance to increase dgraph free tire from 2MB/d ? i want to list you on my free services: cloud.dgraph.io/pricing?type=f… but not sure 2MB a day would help - for most solo devs we run "low-volume projects" (build something, but no traffic) but 2MB/d doesn't help even in that case

Dgraph has been at the forefront of teams building knowledge graphs to power their AI strategy. Today we're excited to announce exciting updates for the future of Dgraph to make that more accessible for teams of all sizes: 🚀 Starting with the Dgraph v25 release in early 2025, we will have a single open-source license (APL) and build for Dgraph with all the features you need to run Dgraph in production 🆕 Serverless Dgraph is now in private preview, creating new, lower-cost on-ramps for teams building knowledge graphs Plus, Dgraph has a new logo! 🔭 Read full story here hypermode.com/blog/the-futur…

Announcing ModusHack ✨👩‍💻 — The hackathon for builders! Join us to create intelligent apps using Modus, AI models, and Hypermode. @hypermodeinc and Hashnode are teaming up to offer $10,000 in prizes for the most innovative projects! From Nov 1 - Nov 30, 2024, it's your chance to win big and build amazing things. Register now 💸

Today is the day! We are 📷 excited to announce Modus—an open source, serverless framework that redefines backend development for AI-ready apps. Built with WebAssembly, Modus combines security, speed, and ease of use for developers!  👇🧵

@codegen trying to reach out but seems the Help form on your site is broken

Learn how to integrate real-time vector search into your application with @HypermodeAI. No need for extensive AI or machine learning expertise—Hypermode makes it quick and simple for anyone to get started. Check out this tutorial to learn more 🤩 youtube.com/watch?v=Me4YjN…

For years, I've heard customers ask to make it easier to deploy and observe their dbt jobs with Airflow. I'm really excited that the day for this has finally come -- customers can now manage dbt and Astro from a single interface.

We'll be open-sourcing @HypermodeAI's "instant vector search" template later this week. In this demo, we run a vector search on a 10k+ item catalog every time the user pauses. p99 response times under 200ms.

Especially at this stage, if you're not deep in both the technical and design details it's almost impossible to make decisions at the pace required. Creating a clear, executable vision is the job and the team we've assembled at @HypermodeAI is delivering!

@TejasKumar_ @cramforce We’re working with a number of companies at @HypermodeAI who have found success with this path. Prototype with the biggest models, then quickly pivot to fine-tune a small, open source model to scale.

@cramforce Thank you, Malte! This is *exactly* what other experts have been telling me. Now to go find resources and do deep dives on how to do this.

I'm building with GPT-4-Turbo and it is incredibly expensive and I cannot realistically scale. To manage costs, this is my best idea: - cache prompts + responses - do similarity search between new prompts and cached prompts - if similar, serve from cache Is there a better way?

@mitchellh Update: just moved to @WarpBuilds for all of our GitHub Actions runners @HypermodeAI. No concurrency limits, better analytics, and remote access to runners. Going to be a tough to sustain differentiation, but excited by what @suryaoruganti and team are building so far.

@mitchellh buildjet.com is doing this, works great to access ARM machines as well

I'm surprised no one has built a niche business around self-hosted GitHub Actions runners. Getting a self-hosted machine is pennies on the dollar compared to the GH pricing and I'd pay a premium to have it turnkey. Am I missing an obvious reason why this doesn't seem to exist? Basic, crude math... macOS: - 6-core macOS on an M1 is $0.16/min ($230/day @ 100% utilization) on GitHub - 8-core macOS on an M2 is $109/month (~$3.60/day @ 100% utilization) on MacStadium - Breakeven point is 681 minutes on GitHub. That isn't much! Linux: - 8-core Linux is $0.032/min ($46/day @ 100%) - 14-core dedicated server on Hetzner is currently $50/month (~$1.60/day @ 100%) - Breakeven point is ~1600 minutes on GitHub. Again, not much. 100% utilization is not unrealistic for large projects with a lot of PRs and long GHA workflows. But even if you get to a fraction of that utilization, there appears to be a huge margin you can profit from.

Soon we'll acknowledge that AI is a tool, not the whole solution.

Despite all the FUD-inducing headlines from vendors, AI applications aren't often built with a single model or technique. A good example here from @TejasKumar_ on combining semantic similarity search and summarization to build a complete experience.

The best founders I know — no matter their company’s scale — thrive on doing customer support directly. There’s literally no better way to understand the pulse of your customer base, what features to build next, or where systems are breaking down. It’s always upside.

@ethanjb @sjmiller609 @gunnarmorling @esammer @DarrenBaldwin03 @JulianLaneve @_marcobambini Risk-based reviews. Nothing more frustrating to an engineer asking for a review on a small fix that passes all tests. In absence of a systematic risk-based approach, compliance has become the justification and the result is weaker reviews where it actually matters.

What’s your magic wand experience here? Curious, @sjmiller609, @gunnarmorling, @esammer, @DarrenBaldwin03, @JulianLaneve, @_marcobambini, @ryanfoxtyler would love to hear your thoughts here…

What is the best GitHub code review UX out there? @kageiit is so right, it’s almost legacy infra, designed for a pre-cloud, release management centric era

@mitchellh Yeah, seems like a natural extension

@ryanfoxtyler No Windows/macOS when that is the major, major pain point I've always had (where the pricing/speed gets stupid). Linux on GHA isn't terrible. Still, happy to see something like this does exist.