Sand2Server

Introducing Cinematic Video Overviews, the next evolution of the NotebookLM Studio. Unlike standard templates, these are powered by a novel combination of our most advanced models to create bespoke, immersive videos from your sources. Rolling out now for Ultra users in English!

It is hard to communicate how much programming has changed due to AI in the last 2 months: not gradually and over time in the "progress as usual" way, but specifically this last December. There are a number of asterisks but imo coding agents basically didn’t work before December and basically work since - the models have significantly higher quality, long-term coherence and tenacity and they can power through large and long tasks, well past enough that it is extremely disruptive to the default programming workflow. Just to give an example, over the weekend I was building a local video analysis dashboard for the cameras of my home so I wrote: “Here is the local IP and username/password of my DGX Spark. Log in, set up ssh keys, set up vLLM, download and bench Qwen3-VL, set up a server endpoint to inference videos, a basic web ui dashboard, test everything, set it up with systemd, record memory notes for yourself and write up a markdown report for me”. The agent went off for ~30 minutes, ran into multiple issues, researched solutions online, resolved them one by one, wrote the code, tested it, debugged it, set up the services, and came back with the report and it was just done. I didn’t touch anything. All of this could easily have been a weekend project just 3 months ago but today it’s something you kick off and forget about for 30 minutes. As a result, programming is becoming unrecognizable. You’re not typing computer code into an editor like the way things were since computers were invented, that era is over. You're spinning up AI agents, giving them tasks *in English* and managing and reviewing their work in parallel. The biggest prize is in figuring out how you can keep ascending the layers of abstraction to set up long-running orchestrator Claws with all of the right tools, memory and instructions that productively manage multiple parallel Code instances for you. The leverage achievable via top tier "agentic engineering" feels very high right now. It’s not perfect, it needs high-level direction, judgement, taste, oversight, iteration and hints and ideas. It works a lot better in some scenarios than others (e.g. especially for tasks that are well-specified and where you can verify/test functionality). The key is to build intuition to decompose the task just right to hand off the parts that work and help out around the edges. But imo, this is nowhere near "business as usual" time in software.

Why the Memory Big 3 Cannot Exercise Implicit Supply Discipline Like HDD Players Someone raised an excellent question: "There are only three DRAM suppliers—why can't they implicitly coordinate supply, like HDD makers do, to avoid over-investment?" Intuitively, it makes sense. In an oligopoly, players should be able to read each other's signals and moderate accordingly. And indeed, after the HDD market consolidated into a three-player structure—Seagate, WD, and Toshiba—the industry maintained relatively stable supply discipline for a considerable period. So why doesn't this work in memory? The core issue is that memory is an industry where those who wait are punished. HDD is a mature industry. The pace of generational technology transitions is slow, and being late to adopt a new generation doesn't get you expelled from the market. This creates room for a tacit agreement along the lines of "let's all take it slow together." Memory is the exact opposite. From DDR4 to DDR5, HBM2e to HBM3, 1-alpha to 1-beta—technology node transitions happen relentlessly, and the first mover to reach a new node captures a 6–12 month premium pricing window. "Exercising discipline" effectively means "falling behind," and falling behind in this industry is fatal. Consider a concrete example. Suppose Samsung decides to conserve CAPEX and delays its transition to the next-generation DRAM specification. But SK Hynix proceeds on schedule. What happens? Samsung gets shut out of socket qualifications on new Intel and AMD server platforms. Socket qualification, once lost, takes months to quarters to reclaim. In other words, the cost savings on CAPEX translate into months of forfeited revenue opportunities. This is the textbook prisoner's dilemma. If all three players collectively restrain capacity additions, everyone benefits. But if even one defects, the other two suffer massive losses. So inevitably, every player arrives at the same conclusion: "I have to move first." On top of this, the cost structure of fabs pours fuel on the fire. A state-of-the-art memory fab generates billions of dollars in annual fixed depreciation. Once a fab is built, it must be run. The marginal cost of producing one additional wafer is extremely low, which means that running at 100% utilization is the only way to minimize average cost per bit—regardless of whether the market is oversupplied or not. HDD factories have relatively modest capital investment requirements and offer flexible production scalability. When orders decline, you simply scale back the line. Semiconductor fabs are a different beast entirely. Shutting one down and restarting it involves enormous costs and lead times, and depreciation charges keep accumulating whether the fab is running or not. While the strategy of "cut production to defend pricing" can work, it comes at the cost of enormous losses. To summarize: the HDD industry features slow technology transitions, high manufacturing flexibility, and a forgiving environment where waiting carries little penalty. The memory industry features rapid technology transitions, enormous fixed-cost burdens, and an unforgiving environment where waiting is existential. Even under the same oligopolistic structure, the fundamental competitive dynamics of each industry are so different that the implicit supply discipline that worked in HDD is structurally impossible in memory.