Clint Sharp @clintsharp

Thanks to @united for another @Starlink flight. And thank god because there is not a movie worth watching. It seems like Elon must be getting all their money because they aren’t spending it on movies.

@adamhjk I admit I’m a bit surprised because I saw this news and I just don’t grok it. Help appreciated with how you’re using it.

I thought this was the best analogy I’ve seen, and he called it three years ago. It’s possible Dario, Sama, Elon, etc believe we need to slow development of AI because they can see things in their labs we cannot. Certainly the Chinese are not going to agree to pace their development. I think David Sacks was spot on. They don’t need anybody’s permission to slow down. They don’t need to have a new regulatory authority make them. They could just do it. Instead, it looks an awful lot like regulatory capture. For companies like mine, much smaller, it’s terrifying there might be a governmental or even industry authority I need to certify our work before we can even ship. As many others have said I think the real story is open models are eating into their growth and the charts they need to show exponential growth have started to bend flatter and they need regulatory capture to lock in their oligopoly. For our users and customers, our experience is the level of intelligence they need for telemetry use cases can be done for pennies on the dollar compared to the boutique token company’s models. It’s an exciting future. Jevons paradox reigns supreme. Lowering the cost of tokens is going to open up so much demand for intelligence. The future is extremely bright, presuming we don’t allow the Baptists or the Bootleggers to lock us out of our path to the market.

Marc Andreessen 🇺🇸 @pmarca

3 years ago

The Baptists And Bootleggers Of AI Economists have observed a longstanding pattern in reform movements of this kind. The actors within movements like these fall into two categories – “Baptists” and “Bootleggers” – drawing on the historical example of the prohibition of alcohol

28 221 1K 362K 561

@kfirgollan Mostly for the last few months I’ve been building apps on top of the product that showcase the power of the product. For the last few weeks I’ve been building something I plan to ship and transition to engineering to run in perpetuity.

How quickly the future becomes the new norm. I'm annoyed that my AI Agent, which I can reach in under 30ms due to the excellent Starlink on my flight, which has generated essentially 3 person months worth of software in the last 24 hours, won't work faster.

I have a use case. I have implemented a coding harness that runs in the isolate, with an emulated filesystem persisted in SQLite. I build React Apps with esbuild, preview them in the browser. Works great, except for a few missing things like npm, eslint, etc. Would be awesome if I could burst out to containers which just run a single tool and report back but still have those containers maintained by celld.

@mattzcarey @pidotdev Me too. It’s awesome. Great foundation for long running agent sessions.

@lmeyerov the results are at secitbench.cribl.io. @nmungel can comment on what we are releasing other than the results I know there’s been a lot of discussion. Some concerns around models training on these datasets etc so don’t know where we landed.

Today, we've released SecIT Bench to evaluate the performance of various models for investigative use cases on top of telemetry. TLDR: open weights models are good enough, especially if we start using LLMs as judges for their output. secitbench.cribl.io

@lmeyerov I didn’t do the work but I’ll ask around about it. I’ve been using pi in my projects.

Blog post with more details on its creation: cribl.io/blog/introduci…

Point an AI agent at your telemetry today and it can only see what one vendor lets it see. Agents are becoming a second consumer of telemetry, right alongside the people who run your systems. Our current siloed systems are not ready for consumers thinking at machine speeds. That’s agentic telemetry, and most enterprises aren’t ready for it. Problem one: agent fatigue. Every vendor is bolting a copilot on its own silo and calling it done. Your security copilot can’t see your application traces. Your monitoring copilot has no idea what your SOC just found. Different agent, same walled garden. Problem two: context starvation. It’s not just missing logs, it’s missing context. A trace without the deploy, an alert without identity, a ticket without telemetry behind it. You can’t ask questions of data you don’t have. Neither can your agents. Problem three: token economics. Your telemetry is measured in petabytes. A context window is measured in megabytes. You cannot shove the former into the latter, and agents ask orders of magnitude more questions than humans ever did. That math breaks fast. The answer isn’t a more features of your siloed tools. It’s an architecture, an AI Platform for Telemetry: many capabilities, integrated, built on open protocols like MCP, where any authorized person or agent can ask questions of all your data, wherever it lives, at full fidelity, with no vendor holding a monopoly on it. Our hope is by defining the architecture openly, you should be able to build this yourself, with @cribl_io or without. Read my blog: cribl.io/blog/what-is-a…

lol, perhaps. The pipeline was always just the start for us. The destination was always analytics. The bigger shift this enables that I didn't think I see you write about is treating this like cloud infrastructure rather than enterprise wide solutions. Customers are provisioning Lakes and Lakehouses as code, and splitting up the workloads to avoid noisy neighbors and better understand spend. This is possible because it is now trivial to federate across multiple repositories where historically that's not how incumbent vendors want you to manage their solutions.

The real reason to distrust Claude: he killed printf('GOT HERE!\n'); RIP.

Perhaps the best thing about building with AI is that it doesn't argue with you

@wadefoster Awesome! I’m in the early days of building something similar for evaluating whether Cribl Investigator can find real environment failures. It’s amazing how fast we can iterate with well defined success function.

You state these like facts but they’re far from certain. From my perspective I don’t see a huge appetite for enterprises to spend more. I also don’t think the limiting factor in the number of companies in the market has ever been the production of the software. The question to answer is how these 10x more companies can solve the problems in a unique and better enough way long enough that the entrenched competitors won’t just copy, giving these new companies the time to achieve anywhere near the scale that would allow most enterprises to consider them a trusted option. Let’s check back in 5 years and see what’s happened.

This just makes absolutely zero sense. There already isn’t a lack of supply in enterprise software. For any given category there are already dozens of choices. There are already funded and bootstrapped in every category. There are free and open sourced alternatives in nearly every category. So what is the limiting factor that drives spend to a handful of category leaders? Trust. Trust is in very short supply. How do I solve this problem for my employer in a way that does not introduce new risk and get me fired? Buyers are not buying just a software package they are transferring risk from them to the vendor. Software is a people business. I’m going on record that a burgeoning supply of dozens of new competitors will do little to change the dynamics of where money flows in enterprise software.

Guru Chahal🇺🇸 @guruchahal

7 months ago

.@nicbstme does a great job capturing the rapidly shifting sands in vertical SaaS! This right here is the key… AI is arming the hordes - and they are coming for their cut of enterprise SaaS spend!

0 0 1 4K 3

Both predictions can't be true: SaaS is dead, categories are going to shrink, and AI will enable tons of new competitors. There will be downward pressure on prices because enterprises are not going to allocate more money to software, but it is very unlikely that categories are suddenly going to flooded with new entrants and buying behaviors are going to suddenly change because of AI. Enterprises already do not opt for smaller competitors, and having dozens more smaller ones attempting to compete for a shrinking pie isn't a path to success for anyone. I'll go on record that the opposite will actually happen. AI will enable software companies to be more efficient. Margins will improve. SBC is dead. It will no longer be a path for a mediocre exec to make millions. But, it will be even harder for startups to enter these categories. The door is shutting and incumbents will be cemented.

@guruchahal Enterprises aren’t looking for more choice. They already pay analyst firms to tell them what to buy so they don’t have to spend time surveying the market. It’s not that nothing will change but the surface level analysis of AI impact on enterprise software is wrong.