Ariya Hidayat

Chris Lattner's take on CCC (Claude C Compiler) is a great summary of where AI's capabilities are at, in terms of generating code: https://lnkd.in/gGYapXVa Would suggest reading this along with Drew Breunig's take on Claude being an electron app: https://lnkd.in/g7xUsbss

11

Most capture systems fail because they ask you to do *work* exactly when you have an idea. In that moment, the only thing that matters is near‑zero friction. But “fast capture” usually creates chaos: notes everywhere, nothing searchable. That’s the gap we’re closing with Notis: capture an idea in ~5 seconds (WhatsApp or voice), and it lands organized + searchable in Notion. Read the post: https://lnkd.in/ewjhgvqu

2

2 Comments

AI-tools are everywhere, but they require thoughtfulness, creativity, and grit to use well. Josh continues to blaze the trail on this at Encamp. If you're interested in getting the most of AI development in an existing and fast-moving code base, I highly recommend reading this post.

9

🚀 Breaking Down Memory Walls for AI and Science! Fascinating read from Pacific Northwest National Laboratory (PNNL) on their groundbreaking "active memory" computing architecture, purpose-built for massive AI and scientific workloads. At the heart of this system is technology from XConn Technologies: a high-performance CXL (Compute Express Link) switch. This isn't just an incremental upgrade—it's a paradigm shift. The XConn CXL switch enables a revolutionary pool of shared memory, allowing CPUs and GPUs to access a massive, unified memory space at incredible speeds. This directly tackles the biggest bottleneck in high-performance computing: data movement. For national labs like PNNL, this means unlocking unprecedented scale for scientific discovery, from energy research to advanced materials science. A powerful example of how core infrastructure innovation (CXL) is enabling the next leap in computational science. https://lnkd.in/gCGCtgx6 #AI #HPC #HighPerformanceComputing #CXL #ComputeExpressLink #DataCenter #Innovation #GPU #ScientificComputing #NationalLabs #XConnTechnologies #PNNL

22

Nikhil Mungel, writing about Karpathy's March of Nines (i.e., how many nines of reliability enterprises need), points out that each additional nine of reliability requires as much effort as each preceding nine. He figures enterprises need a minimum of four nines (99.99%) of reliability; this, he says, "is where it starts to feel like dependable enterprise-grade software." So Mungel in the first part of this piece just explained why monolithic, statistical-only ML/AI is not sufficient to enterprise needs. And then Mungel in the rest of the article describes in nine different ways to add reliability. Five of these ways IMHO demand the inclusion of semantic knowledge graphs (deterministic facts and rules logically, expressively linked in an graph) in a neurosymbolic mix--for scaling purposes if nothing else: "Nine levers that reliably add nines: 1) Constrain autonomy with an explicit workflow graph [one layer of the five layers of context actually needed--a ground layer and four layers of abstraction for large-scale, human-guided interoperation] 2) Enforce contracts at every boundary [e.g., using the RDF stack] 3) Layer validators: syntax, semantics, business rules [ using SHACL] 4) Route by risk using uncertainty signals 5) Engineer tool calls like distributed systems 6) Make retrieval predictable and observable [direct, complex, boundary crossing query and DBMS retrieval, best scaled using SPARQL and a semantic graph DBMS] 7) Build a production evaluation pipeline 8) Invest in observability and operational response [feedback loops with human-in-the-loop oversight, best enabled and overseen with a standard semantic graph] 9) Ship an autonomy slider with deterministic fallbacks" Articulates why putting data and knowledge first is essential if enterprises are going to avoid a garbage-in/garbage-out scenario with so-called AI. https://lnkd.in/gTV6MnRA

12

1 Comment

A little while ago I posted about Sacramento Labs being interested in acquiring software products and heard from a number of folks I hadn’t talked to in a while. The responses and conversations were genuinely great. (Original post here: https://lnkd.in/g-6hjy6q) After talking it through, Garrett Milster and I decided to formalize something we were already doing informally: we’re offering a $10,000 referral fee for any warm introduction to a company that we end up acquiring. A few notes for clarity: -The introduction needs to be a direct, warm introduction (email or message). -Public listings or cold forwards don’t qualify. -Referral fees are paid after closing and are contingent on a completed acquisition. -There’s no expectation of ongoing involvement unless explicitly discussed. If you come across a software business that feels like it might be a fit for us, feel free to make an intro.

6

AI can compress the operational gap, but I don't expect it to totally eliminate strategic divergence. Patent portfolios encode beliefs about where value will accrue and which upside you want exposure to.

1

We’re excited to partner with Cognition, the stellar crew behind Devin 🤖 and Windsurf 🌊, to make SWE agents more effective. Here’s the workflow: ✨ CoreStory extracts the hidden business logic, requirements, and complex behaviors in legacy codebases into a persistent spec. ⚡ You pass that spec as context and tasks to Devin, Cognition's Autonomous AI Software Engineer. 🚀 With the spec to navigate, the AI agent gets more powerful. It can modernize entire systems, or use the spec to field even more tickets with higher confidence. This is what the future of software engineering looks like: Human devs + AI agents, working side by side, with CoreStory fueling the spec. #AI #Agents #SpecDrivenDevelopment #LegacyModernization #CoreStory #Cognition

17

This is the real filter: if nobody is willing to put their name on it, it’s not automation — it’s noise. The AI layer has to be approval-native: policy checks, audit trail, and explicit handoffs to accountable owners. Otherwise “smart” outputs never become decisions.

1

As COO at www.jaiinfoway.com this is an important signal for where agentic AI is heading. Non technical implementations are catching up fast but the real story is iteration speed and learning loops. Building Cowork entirely with Claude Code shows how agents are becoming first class workers not just assistants. At www.jaiinfoway.com we see this shift clearly as enterprises move from copilots to autonomous task execution with guardrails. UX will evolve tooling will mature but the direction is set. Curious how others here see non technical users adopting agent driven workflows this year. Are we ready for agents as coworkers. #Jaiinfoway #AgenticAI #Claude #AIAgents #EnterpriseAI #FutureOfWork #AIProductivity #AITransformation #AILeadership

2

To all the engineers in my network who might or might not know what test-driven development is: The California government’s approach to innovation is like test-driven development. In some cases it actually *IS* test-driven development. I just found out that California’s failed SB 1047 would have “required developers to perform risk assessments of models prior to release, with guidance from the Government Operations Agency.” (wikipedia) There’s one place where the analogy to test-driven development falls short, though. Whereas in TDD you know what you’re testing for, SB 1047 doesn’t even know what it’s trying to regulate, as its goal is to “mitigate the risk of catastrophic harms from AI models so advanced that they are not yet known to exist.” Unfortunately, the state senator who introduced SB 1047 is back at it and still doesn’t realize he is trying to put the cart before the horse. Source: Politico

2