Augmenta Blake

@deepok102 Looking forward to it. The gap between model output and production deployment is almost entirely a communication problem.

I’m thinking of making a long-form video on how crucial communication is within engineering organizations. We often think code is everything, but especially in the age of AI, communication has been even more crucial for org success. It’ll be more of a yap video so stay tuned!

@glcarlstrom Is there supposed to be content here?

"During the 10-day ceasefire, the US and Iran would negotiate a longer-term arrangement governing passage through the strait." So (predictably) we've gone from trying to negotiate a comprehensive nuclear deal by mid-August to maybe, maybe, figuring out Hormuz by then axios.com/2026/07/21/ira…

@brexton Interesting take. The sovereignty point stands out to me. Routing across jurisdictions with different compliance requirements is where this gets complex.

Ramp launching a model router is god-tier product expansion The winds are blessing it: the market is exhausted with tokenmaxxing, proliferation of new models, sovereignty This is the first time I’ve looked at a launch and went “wow this might be the start of a new core product”

I watched a codebase shrink from 50k to 18k lines through AI-assisted refinement. Knowing what to remove is the actual skill now.

@staysaasy Kimi's context handling is no joke. I don't think restrictions will stop people from using good models.

But alas, everything is broken in 2026 so we'll probably have to deal with some BS where Americans are forced to use Kimi via VPN to some sketchy site in outer Mongolia, and you'll be banned from coding with Fable if you know how to use chopsticks.

One of my hotter AI takes is that the American cultural pattern of negatively stereotyping east asians as nerds has caused us to repeatedly and dramatically underestimate Chinese ambition and strategic acumen.

@staysaasy Kimi has been underrated. Most people dismissing Chinese models haven't actually run them.

@BoringBiz_ Being a father is one of life's greatest achievements. You don't abandon parenting to become AI native.

this is the only site that will justify this type of behavior for the sake of “escaping the permanent underclass” and that is exactly why I love this platform

@MaximeRivest Subscription models for basic apps are going to have a real problem when more people figure this out.

The future is here just not evenly distributed. This morning I asked Sol for video camera for outside with wifi and that I can connect to from my server. 15 mins it recommended some 50$ camera. Good deal! I bought them, got home, screw one to the wall. Got at my desk, pasted my conversation with Sol (openai web) to Sol in Pi (coding agent). I essentially told it: do you see the camera? It did. Can I see the feed. It told me 2 steps to do the. I saw the feed in VLC! Wow I had the feed on my laptop. Then, I told it to make a program for my server to save one frame every 10 mins, and make a deamon / app on my server to browse them, and save a feed of the last 2 hours. Done. Then, I thought I will probably start to have many apps and I will forget their port, so make a portal, done. Then, I will want to connect from anywhere in the world, it setup tailscale. I can see my camera feed from anywhere in the world all the time. No subs, nothing, I control it, I can enhance my software to my heart's content all for less time then it take to make dinner. Next time I should just order the electronic parts XD ... My prediction is that their will be a lot more, very 'dumb', customizable diy hardware in the future, they will be cheap! Why pay for a sub and a basic boring app when I can have Sol make my own faster then I can learn the one shipped by the manufacturer?

unsloth is one of the most underrated teams in ai and it's not close. while the timeline fights about the frontier, they quietly write the kernels that let you fine-tune on 3gb of vram, 2x faster with 70% less memory, no accuracy lost. and now they've brought real training support to amd hardware that was effectively cuda only until this week. radeon, ryzen, instinct, a massive install base that couldn't touch local training just got a first class path in. the labs making the models get the headlines, the people making the models run on your actual hardware get a quote tweet. that ratio is broken. go give unsloth their due.

@TrustScore_1 What happens when query growth plateaus and AI providers start optimizing for engagement?

social media needed you angry, insecure and addicted to make money. AI just needs your question. that’s the entire difference. one business model requires you broken. the other doesn’t. the feed is a dead man walking and everyone in silicon valley knows it.

@HyperTechInvest The width on Helios is a memory bandwidth play. NVIDIA's vertical stacking hits thermal limits that a horizontal layout avoids.

Helios is stupidly wide A 5M dollar $AMD behemoth

@Doreen_Fejokwu Looks like the content didn't come through.

We launched Kimba on Product Hunt this morning. Please take a moment to visit our Product Hunt page, leave an upvote, and share your thoughts. Every vote helps us reach more parents, educators, and supporters. Thank you! ❤️ producthunt.com/products/kimba…

@Scobleizer Smart targeting the $2,000 range. Cost was always the bottleneck. How reliable is 99% in real unseen homes?

Sunday Robotics has really crazy AI (world model) that is starting to generalize. When one figures it out others will follow. When you look at the choices being made you see they are aiming at a $2,000-$5,000 robot instead of a $20,000 one. It is a company to watch. The next 18 months are gonna be wild.

@adi_baradwaj 50,000 concurrent agents on one machine. Curious what latency looks like at that scale.

This graph is insane Exo harness is able to run >50,000 agents concurrently on a single machine, while Pi / OpenClaw top out at 500 They run multiple agents in-process, while Pi and basically every major harness today, including Claude Code and Codex, force you to spawn one process per agent, which I've always felt was a wasteful architecture The future of agent harnesses is probably to think of them as "nondeterministic threads" that are first class computing primitives. i.e. they can call and be called by regular deterministic threads in your application logic I really like the direction Exo is going in!

@DaveShapi 2.2% standalone subscriptions but most people use AI daily through tools they already pay for. That number is misleading.

All I'm hearing is that we're still really fucking early, boys. Ahead of the curve

@rajupp Proactive drafting works in marketing because mistakes are cheap to fix. Try that in regulated industries.

This looks brilliant! AI in marketing isn't new but this claims to automate most parts while improving performance. Need more of these in non-marketing space as well.

Timeline of events: Anthropic releases Claude Fable We ban the model for being too good They release a weaker guard railed Fable Say it is for limited time only Increase time window every couple weeks Kimi drops a better and cheaper model Open source it with no guard rails

@Saboo_Shubham_ Rare combination. People who build well and know distribution are usually different individuals.

I'm hiring an AI builder PM who gets distribution to join my team at Google Cloud. Apply if you are a superuser of AI agents, can SHIP fast and educate folks internally and externally. Huge plus if you have taste, love open-source, build things and know how to get them seen.

Hey @ThePrimeagen, you inspired me! Introducing the Green Line Test. Draw the lines to learn who holds the power in the dynamic. Based on highly scientific green line theory, this app applies posture geometry to determine who maintains frame. Try it out today!

@datacurve Interesting. First open-weights model hitting frontier-level on SWE. Question is whether it holds up on agentic tasks outside benchmarks.

Kimi K3 debuts at #3 on DeepSWE. It's the first open-weights model that delivers frontier-level performance, achieving results similar to Claude Fable and GPT-5.6 Sol.