Is this you? As a journalist, you can create a free Muck Rack account to customize your profile, list your contact preferences, and upload a portfolio of your best work.
Claim your profile
Articles by Jonathan Murray
Jev by TypeSafe on Backboard.io
No fancy long posts here, just cold hard facts. The viral model Jev, is now on backboard: HUGE ANNOUNCEMENT: Backboard.io now supports TypeSafe Jev — System One Models • Added Jev support for structured decisions: Choice, Score, and Noul (true/false probability). • Use the existing message API with llm_provider="typesafe" and system_one.questions. • Evaluate text or structured data, with conversation history included.
Make a Runway Clone in one shot, or something else!
i'm jon, i co-founded backboard.io. this is a launch post. it's short and there's code. video models are live on backboard. Here is a video of me and my dog rapping if you don't believe me. ByteDance's Seedance family is in, served through OpenRouter, along with the rest of the OpenRouter video catalog. that's the news. here's why you should care. video is the worst part of the stack to glue text you've figured out. images, mostly. video is where it gets ugly. every provider has its own job api. submit.
OpenAI can't rule out that it stole its most recent breakthrough
A mathematician spent a year on one of the hardest open problems in math. He asked OpenAI a simple question. Did you train on my sessions? Today they answered. Sort of. Here is what happened. The setup Tristan Buckmaster is a math professor at NYU. He and Levent Alpöge spent most of the past year attacking finite-time blowup for fluid equations, the family of problems that includes the Navier-Stokes Millennium Prize. They worked with LLMs the whole way. Claude, Codex, GPT-5.6 Sol, Astra.
There Are Only Two Ways to Make Money in Software.
my team builds AI infrastructure, so yes, i have a horse in this race. the horse shows up at the end, clearly labeled. the history comes first, and the history is free. jim barksdale, back when he ran netscape, said the only two ways to make money in software are bundling and unbundling. people quote it like a joke. it's not a joke. it's the whole industry in one sentence, and it explains exactly what happened to your sprint board.
I Watched Our Interns Play Valorant for Three Hours. Put Your Rank on the Resume.
Last week of summer with our interns we skipped the goodbye lunch and booked an esports arena. Split them into teams. Valorant. Best of whatever until the room kicked us out. I run go to market at a startup. I have sat through a lot of team exercises. Trust falls. Escape rooms. A guy with a flip chart. None of them came close to what I watched for three hours in that room. The quiet ones were calling rotations. The loud ones shut up and listened.
The Matrix Wasn't A Battery Farm. It Was A GPU Cluster Made Of Human Brains.
Nvidia is worth more than most countries because we cannot figure out how to do cheap inference. There are eight billion units walking around that do it on 20 watts. I rewatched The Matrix recently and got stuck on the same thing everyone gets stuck on. Batteries. Morpheus holds up a Duracell and tells Neo the machines are farming humans for electricity. Bio-electricity, body heat, 25,000 BTUs, the whole speech. It's a great scene. It's also thermodynamic nonsense.
Every Layer of Your AI Stack Is an Attack Vector. Count Them.
From the team at Backboard.io. We build AI infrastructure, so we have a position here. We state it at the end, clearly labeled. Everything before that is just counting. TL;DR A typical production AI application runs on 6 to 9 separate vendors: a model gateway, a vector database, a memory service, a RAG framework, an embedding provider, an orchestration layer, an observability tool, and 2 to 3 model providers.
I Counted the Attack Vectors in Our AI Stack and Now I Can't Sleep
so a while back i did something i do not recommend before bed. i counted the vendors in a normal production AI stack. not a weird stack. not a bad team's stack. the stack everyone builds, the one in every architecture diagram on this site. model gateway. vector database. memory service. rag framework. embeddings provider. orchestration. observability. two or three model providers. that's eight. for one app. and here's the first thing that got me: nobody approved that stack.
Anyone ever go back to their notetaker outputs and summaries?
Anyone ever go back to their notetaker outputs and summaries?
Nash Is Now Open Source; Meet Your New AI Workspace
Big news: Nash is now open source. You can find the repo here ğ github.com/Backboard-io/nash Nash started as an internal tool to make AI actually usable day-to-day â not a chat window bolted onto a model, but a real workspace. As of this release, it's open for anyone to use, fork, self-host, contribute to, or build on top of. Alongside open-sourcing the project, we're also shipping a major release that pulls Nash's features into a single, connected experience. Here's what's new.
Benchmarks Don't Build Great Products. Engineers Do.
The AI industry has a benchmark problem. Not because we have too many benchmarks. Because too many companies treat them like trophies instead of tools. A benchmark's primary job is to make your product better. Publishing the score is secondary. The test I use: a benchmark should challenge your engineers before it impresses your marketing team. If it isn't making your product better, it probably isn't serving its most important purpose. Why does the AI industry have a benchmark problem?
How to Let OpenRouter Pick Your Model, and Choose Who Serves It
Backboard now supports OpenRouter automatic model selection (openrouter/auto) and per request provider selection. Here is how both work, with JSON examples for the Backboard API. Short answer Backboard now gives you two independent routing controls for OpenRouter requests: Automatic model selection. Set llm_provider to openrouter and model_name to openrouter/auto. OpenRouter classifies the prompt and picks the model. You pay the standard rate of the model it chooses, with no additional router fee.
Coding agents got boring the moment we built a really good one.
MESSAGE FROM BACKBOARD.IO Co-Founder Rob Imbeault: Coding agents got boring the moment we built a really good one. That sounds dismissive. I’m ok with that. Cursor is worth tens of billions. Coding agents are attracting enormous amounts of capital. The category is clearly valuable. But building R-CLI changed how I think about where the value actually sits. We built one of the best coding harnesses in the world in a matter of months. On Terminal-Bench 2.1, it performs at the frontier.
We Open Sourced R-CLI, the Coding Harness Above Every Published Terminal Bench 2.1 Result
R-CLI is now open source. MIT license. The whole harness. Repo: https://github.com/Backboard-io/Backboard-R-CLI First, the ask We are giving away the harness behind the highest scores ever recorded on Terminal Bench 2.1. Free, MIT, running on your own keys in about two minutes. The receipts are below, with public verifier logs for every claim. The entire price: Star the repo. Stars are how open source projects get found. Thirty seconds. Share it with one developer who pays too much for AI coding.
We hit 99.95% on the LoCoMo memory benchmark. Here's the catch, and why it still matters.
Our CEO Rob Imbeault published a piece on LinkedIn this week about a result our team posted: 99.95% on LoCoMo, the most cited benchmark for long-term conversational AI memory. His version is written for executives and CISOs. This is the developer version, because the interesting part isn't the number. It's how we got it, and what it says about where memory architecture is heading. (Credit where it's due: the ideas here are from Rob's original article.
Our small team just outscored Claude Code and Codex on Terminal-Bench
I want to be upfront about something: I didn't build this. Our engineering team did. But I get to write about it, and I've been waiting a while to write this one. This week we submitted the Backboard CLI to the official Terminal-Bench 2.1 leaderboard. The score: 85.4% ± 0.8%, with a pass@5 of 0.888, running Claude Opus 4.8 via Bedrock. For context, the top published entries on the leaderboard right now are Claude Code with Fable 5 at 83.8% and Codex with GPT-5.5 at 83.1%.
To big to lie, cheat, steal?
Remember when Volkswagen cheated and lied on their emissions testing for their clean diesel… That was so crazy in retrospect. Imagine Anthropic and OpenAI were completely training on everyone’s data and just lying about it and promising not to in their Terms and Conditions? I guess we’ll see.
Apple wins the AI race they're not even in.
Apple is going to win the AI race. Not because of a model. LLMs are not the alpha. Here's what nobody is pricing in. Lease a $2,999 MacBook Pro. $58 a month for 36 months. Then you owe $911 to own it or you hand it back. Apple gives you 6 more months to decide. You keep paying $58. Now you have 6 months to find $563. Can't do it? Return the Mac. Do the math. 42 payments of $58 is $2,436. Add the $563 balloon. That's $2,999. MSRP. Not a dollar more. No interest. No lease fee. No big cheque up front.
Dario must be crashing out
We’ve decided to Open Source both HelloNash.ai as well as Backboard Studio in solidarity of Jensen and the AI community. Backboard is here for Sovereign AI, and so that you can own your intelligence. Your business is not our business.
am I on the hook for a Private Jet?
So a few months ago my co-founder Rob stood up in a meeting and told our interns that if they got Backboard to the #1 spot on Terminal-Bench 2.1 we would rent a private jet and take the whole team somewhere to celebrate. It wasn't a joke. It was a challenge. It was just an insane one. Terminal-Bench is the most sought after eval in AI right now, the thing every trillion dollar frontier lab is openly gunning for, and Rob told a group of interns to go beat all of them.
The team is unreal.
Are you sure you want to hide this comment? It will become hidden in your post, but will still be visible via the comment's permalink.
We bet against the GPU arms race. Here's what shipped.
On July 1 we announced four things at once. The press release version is here. This is the version for people who actually build things. The short story: while the industry spends hundreds of billions on new hardware, we took the opposite bet. Get more out of the GPUs that already exist, and keep everything inside the customer's own environment. Here's what came out of it.
The Prophet and the Price Cut
Two things happened this month and they tell you everything about where AI is actually going. Coinbase quietly cut its AI bill nearly in half. Open models, smarter routing, better caching. No drama. A finance footnote that happens to be a glimpse of the future. And Dario Amodei published another essay. Not a tweet. An essay. The kind of sprawling, twenty-thousand-word civilizational scripture he keeps handing down from the mount.
Banger from one of the squad πͺπ»π
Banger from one of the squad πͺπ»π
Internmaxxing vs. Old Man Shakes Fist at Cloud
Internmaxxing Somebody on your timeline this week called intern code "API slop." Confident. A little smug. Probably got a few hundred likes. It's bullshit. I run a cohort. About half my team is interns right now. I gave them equity options ahead of schedule because I'm not interested in pretending the work is junior when it isn't. The projects shipping out of that group are robust. Production worthy.
Why the Fable 5 Crisis Proves Your AI Context Layer Can't Live Inside the Model
On Friday, a single government letter pulled a frontier AI model off the internet for everyone. The Commerce Department issued an export-control directive on Anthropic's Claude Fable 5 — the reported concern being that its guardrails could be jailbroken. To comply, Anthropic disabled it for all customers, not just foreign nationals. One letter at 5:21pm on a Friday, and every developer and team building on Fable 5 woke up to nothing. No deprecation notice. No migration window. Just gone.
We built a coding harness that beats frontier models using open ones. It's in open beta.
Here is the bet we made: build software memory-first, not model-first, and it will outperform. Everyone else is racing to wrap the next model. We did the opposite. We built the memory layer first, the routing first, tool-calling, now the recursive engine, then let the model be a swappable part. Today that bet has a name: Backboard Development Studio. It starts with the R-CLI, a coding harness now in open beta. The headline result? It beats frontier models using open ones.
our #1 user of Backboard.io is a memory company positioning themselves as a competitor to us, white labeling our product...
our #1 user of Backboard.io is a memory company positioning themselves as a competitor to us, white labeling our product...
We're still the only one to hit #1 on both LoCoMo and LongMemEval. Here is how to use it.
Backboard is #1 on LoCoMo and LongMemEval, the two academic benchmarks for long-term AI memory without changing the original guidelines. Other companies have gamed by using newer models with bigger context windows. This post explains why the result matters anyway, what it actually measures, and how to use the memory that earned it. What these benchmarks test These are not "find a fact in a wall of text" tests.
Chat with your documents: agentic RAG in a few lines
RAG is not new. Chunk a document, embed the chunks, store them in a vector database, run a retrieval step on each query, then feed the results to the model. Every team building with AI has wired this up at least once. It works, and it is also a stack of moving parts you have to assemble and keep running: a parser, an embedding model, a vector store, a retriever, and the glue between them. Backboard does nothing novel here. It just puts the whole thing behind one API.
Your memory, your data: read, edit, export, delete
Most AI memory features are a black box. The assistant remembers things about your users, but you cannot see what it stored, you cannot fix a wrong fact, and you definitely cannot take the data with you if you leave. Your users' information lives in someone else's system on someone else's terms. We want you to be here by choice, not by force. Backboard treats memory as your data. Every memory an assistant holds is readable, editable, exportable, and deletable through the API. No black box.
Trying not to spam but there so much to sharrrrreeeeeeeeee
Trying not to spam but there so much to sharrrrreeeeeeeeee
Give your AI memory in one parameter
By default, an LLM forgets you the moment a conversation ends. Start a new chat and it has no idea who you are, what you told it last week, or what you prefer. For a real product, that is a dealbreaker. Users expect the app to remember. The standard fix is a memory pipeline you build yourself. Extract the important facts from each conversation. Turn them into embeddings. Store the vectors in a database.
Stop letting your hackathon API keys rot
You've got OpenAI, Anthropic, Gemini, and xAI credits sitting in five dashboards. Plug them all into one API and get free state management, courtesy of Dev.to and MLH. If you've done a hackathon or run a startup, you have API credits scattered everywhere. OpenAI from one event. Anthropic from another. Gemini and xAI from your last sprint. All sitting in separate dashboards, half-used, slowly expiring. Backboard fixes that. One API, your keys, every model.
If there was an alternative to Claude Code, Cursor, Codex, Factory, etc. that performed better on benchmarks, consumed up to 30% less tokens, cost up to 90% less and DID NOT train on your code and data, would you switch?
If there was an alternative to Claude Code, Cursor, Codex, Factory, etc. that performed better on benchmarks, consumed up to 30% less tokens, cost up to 90% less and DID NOT train on your code and data, would you switch?
Stateful AI without a database: threads and assistants
LLMs are stateless. Every API call to a raw model is a blank slate. The model has no idea what was said two messages ago. So the moment you want a chatbot that remembers the conversation, you are on the hook for state. The usual answer is infrastructure. Spin up Postgres to store message history. Add Redis to cache sessions. Stand up a vector database for long-term memory. Write the code that loads history, trims it to fit the context window, stitches it into every prompt, and saves the new turn.
Send your first AI message in one API call
Most AI tutorials start with a setup checklist. Pick a model provider. Create an account. Wire up a vector database for memory. Stand up a server to hold conversation state. Glue it all together. Then, finally, you send a message. Backboard skips all of that. One API call sends your first message. A thread, an assistant, memory, and routing across thousands of models are already running behind that single call. You do not assemble the stack. It is the stack. Here is the whole thing.
Toward a Standard Model for Agent Memory
Most agent memory systems are digital attics. You put things in. You hope to find them later. You mostly don't. The retrieval is fuzzy, the context is lost, and the agent that needs to remember why a deployment failed three weeks ago gets back something that looks related but carries none of the causal weight. This is the wrong mental model for memory. Not because the retrieval is bad — though it often is but because storage is the wrong frame entirely.
If Microsoft and Uber can't afford AI coding, what chance do the rest of us have?
Two stories landed in the same news cycle. Microsoft cancelled most internal Claude Code licenses. Windows, Surface, Teams, Outlook, all migrating to GitHub Copilot CLI by June 30. Reporting is consistent on the why: usage exploded, the bills got indefensible, and the company that owns Azure and is one of Anthropic's biggest partners decided it was cheaper to migrate thousands of engineers than to keep paying the meter.
OpenAI and Anthropic are Friendster and MySpace, if Subquadratic proves to be true.
If you've ever shipped an LLM-powered feature that needed to reason over a real codebase, a real contract, or a real research corpus, you already know the shape of the problem. The model technically accepts a million tokens of context. In practice, the answers get worse as the context gets longer, and your infra bill gets worse faster than that.
Very cool use of Backboard!
Very cool use of Backboard!
"Of Course" Erodes Trust Faster Than Bad Code ... Two Words That Are Killing Your Career
You already have the job or the internship. You're on the team. You're in the meetings. You're in the Slack channels. And the thing that's going to hold you back has nothing to do with your code. Someone you work with says "hey, could we do X?" and you say "yeah, of course." Feels confident. Feels like you just proved you've got it. But you gave an answer that contains zero information. No cost. No timeline. No tradeoff. No indication of whether you even understood the question.
Why LLM Reasoning Is Breaking AI Infrastructure (And How to Fix It)
If you've tried building anything serious on top of large language models (LLMs) recently, you've probably run into this: "Thinking" is supposed to make models better. In practice, it makes your infrastructure worse. This isn't a model problem—it's an infrastructure and abstraction problem. And it's getting worse as teams scale across multiple AI providers. Let's break down exactly where things go wrong.
The Hidden Challenge of Multi-LLM Context Management
Building AI products that span multiple LLM providers involves a challenge most developers don't anticipate until they hit it: context windows are not interoperable. On the surface, managing context in a multi-LLM system seems straightforward. You track how long conversations get, trim when needed, and move on.
I Broke SSO Trying to Center a Div. Let's Talk About Tokenmaxxing
A couple weeks ago, I tried to recenter some text on one of my side project SSO pages. That's the whole task. Move the text. Left a bit. Right a bit. Until it's in the middle. Center. Middle. I opened Claude Code. I said, roughly, "hey, center this." Fifteen minutes later I was two bugs deep, SSO was broken — not the text, the whole login flow — and I'd hit my usage limit trying to unbreak the thing I broke while trying to do the thing that should've taken eight seconds in the inspector. Palm. Face.
Stop building what your customers ask for
I was at a conference this week. Bunch of stakeholders on stage. Hospital admins, big-name buyers, a couple of policy folks. The message to founders was loud and clear: "You need to be consulting us. You need to be adapting your products to our suggestions." And honestly? I hated it. Not because they were completely wrong. They were half right. They were just shouting the half that was wrong. Here's the part that's true Building in a vacuum is how you ship things nobody uses.
ππ"ππ"ππ"ππ"πͺπ"πͺπ"πͺπ"πͺπ" - DEV Community
ππ»ππ»ππ»ππ»πͺπ»πͺπ»πͺπ»πͺπ»
My co-founder is just being honest in this post. ;)
My co-founder is just being honest in this post. ;)
Two Days, Two Hacks: The Lovable Disclosure and the Pattern Nobody Wants to Talk About
Yesterday I wrote about the Vercel incident and walked through why you need to rotate your "non-sensitive" environment variables today. I thought that would be the week's security post. Then I woke up to @weezerOSINT's disclosure about Lovable, and now I am starting to wonder if someone out there is just running an end-to-end test on the mythos of the modern AI-dev stack. Two days. Two incidents. Totally different root causes. Same uncomfortable conclusion.
I Think Therefore I Am… A Big Pain in the A$$
If you’ve tried to build anything serious on top of LLMs recently, you’ve probably run into this: “Thinking” is supposed to make models better. In practice, it makes your infrastructure worse. Let’s break down where it actually hurts. The illusion of “just turn on reasoning” At a high level, you’d expect something simple: Turn reasoning on → better answers Turn reasoning off → cheaper, faster Reality is messier.
Show More
loading
Actions
Is this you?
As a journalist, you can create a free Muck Rack account to customize your profile, list your contact preferences, and upload a portfolio of your best work.Get in touch with Jonathan
Contact Jonathan, search articles and posts on X, monitor coverage, and track replies from one place.
Learn more about Muck Rack