San Francisco, CA
Joined January 2019
- Tweets87
- Following137
- Followers30
- Likes26
Pinned Tweet
Recording of our #CriblCon 26 talk just went up if you're interested in learning the engineering principles behind our app framework
With the growth of agentic development, platforms running untrusted workloads require sandboxing and strict trust boundaries.
Broke down the architecture and primitives behind the new Cribl App Framework today at #CriblCon 26 with Glenn
1
113
Great read from Cloudflare. It also shows why we didn't go this route at Cribl.
It's hard to run well for teams with experience. Now imagine shipping this to on-prem customers.
We opt'd for a simpler approach to just keep collection jobs running when the leader node crashes
We're excited to announce that Cloudflare is building an experimental consensus service using a never-before-deployed consensus algorithm, QuePaxa. cfl.re/48bLI84
1
43
Spinning up my own scaffolding from the reference architecture to see if these memory structures and dreaming across all repos improves the efficacy of my sessions. Curious if the offline consolidation moves the needle here. Will instrument the shit out of it.
Report back soon
Memory is what allows Claude to learn from experience. In a new Engineering Blog post, we share how “dreaming” helps Claude organize what it’s learned, correct its mistakes, and find what it needs more easily the next time. anthropic.com/research/compo…
2
197
Don't fall for the rage bait... Don't fall for the rage bait...
It's not bad engineers, it's lack of context. It's explicit abstractions. It's preventing breaking changes during refactors. It's avoiding misuse of the language before runtime.
Damn it...
I was never a fan of Typescript.
I never had bugs caused by typing (or lack thereof) in JavaScript because I know how to write code.
TypeScript was created because of bad engineers. We built a whole language to protect ourselves from them.
Now we don't have bad engineers anymore, we have bad LLMs.
We really only need runtime validators now.
43
This is where I see the SaaSpocalypse playing out.
I'm curious how this fully plays out for consumer tech, but enterprise tech will likely start moving this way.
Customers will begin demanding single tenant, if not on-prem architectures to protect themselves
I think @Tailscale could be 100x larger than it is today in just a few years.
The open internet is dying. Agents will turn it into a toxic wasteland. UniFi, Tailscale, Discord, and AWS all win in that scenario.
Call of Duty lobbies are a good analogy. When games are overrun by cheaters (humanoids), humans retreat to private, high-trust lobbies.
The same thing happened with the Facebook Wall (blast from the past): people retreated to group chats.
Besides not being a cozy place to hang out it’s also dangerous to advertise your existence (dark forest theory of the internet)…
🔐🕸️
1
3
454
Software engineering != Software development
Engineering was never about writing code. That was the means to the end.
Engineering is using judgment to pick the right tradeoffs to solve a problem given a set of constraints.
Construction is the commodity. Judgment isn't.
Restaurants also have razor thin margins. SaaS isn't dead, but the business model and the way the software is built must be rethought. Your SaaS will have to cost less than building and operating your own with agents
"Saas is dead because customers can vibe-code their own tools."
People can cook. Restaurants still exist.
1
2
41
AI writing code impossibly fast is awesome. I can build anything immediately. But that used to be the fun part. AI spinning up infra for me in AWS to POC something, running that test, and tearing it all down within seconds is massive. Fair trade.
38
The biggest failure case I'm seeing with AI is the lack of technical leadership capacity. Slow construction enabled architects to digest and course correct architectural violations. Now the architecture can be replaced in an afternoon without anyone knowing
1
56
If you're defaulting to filing a ticket for a human to triage, your engineering org has a bug. I'm not a "death to PRs" type, but I'm 100% "death to tickets".
Not sure how jira survives the AI era
74
Every dev is one line in a markdown file from being the greatest developer on earth. Trust me bro.
68
I'm embarrassed to admit how amazing I find it that slack now natively renders my LLM-produced html report/presentation
55
Nick Romito retweeted
Introducing cribl-decision-1.0!
Every telemetry event is a decision: where to route it, how to enrich it, and what needs attention. Today, we're releasing a 4B parameter, open-weight model that returns probabilities for your supplied choices in a single forward pass - a foundation for bringing AI decisions into telemetry pipelines.
The weights are available on huggingface: huggingface.co/cribl-ai/crib…
Read more: cribl.io/blog/cribl-decision…
Was nice to have independent js runtimes innovating, so @nodejs could borrow the ideas. Curious how it goes now that deno and bun have both been snatched up by big tech
We are excited to announce that Deno is joining Cloudflare!
deno.com/blog/cloudflare
54
Markdown prompt files feel like the new bash. A tiny file solves a problem quickly. Then someone adds another line. And another. Suddenly you have an unmaintainable mess where a change subtly breaks 5 different things.
At least bash broke deterministically...
1
1
58
Unit tests being viewed as bloat because "AI gets it right the first time" is a weird take. I mean, bad unit tests were always bloat. But good tests ensure abstractions and correctness over time, not just the first time you build something.
2
44
With the growth of agentic development, platforms running untrusted workloads require sandboxing and strict trust boundaries.
Broke down the architecture and primitives behind the new Cribl App Framework today at #CriblCon 26 with Glenn