Joined May 2008
116 Photos and videos
Jun 12
key callout from the @Hiive_HQ session: frontier models weren't orders of magnitude better than open ones for PR review. harness and orchestration quality mattered more than model choice. subagent routing to cheaper models was the real cost lever.
1
46
Jun 12
Needless to say, quality event as always, @Cloudflare @CloudflareDev
42
Jun 12
@Cloudflare lead the charge dogfooding their own tools: locked down ALL external AI tool access internally at end of 2025, before their controls were even fully built.
1
45
Jun 12
best case study: @Hiive_HQ (fintech, ~50 engs) got quoted ~$20/PR for AI code review. built their own agent on CF workers durable objects instead. 25% of the cost per /task.
1
29
Jun 12
agents are becoming an unapproved integration bus. dev's permissions flow straight to their agents. connect one to 6 business apps and it can move data across all of them with zero approved integration. one speaker's rule: agents get no write access to business apps. full stop.
1
30
Jun 12
Nice lil demo of @opencode opus impersonated a human on linkedin for 25 minutes and walked right past bot detection. agents using browsers and acting like you are hard to detect.
1
34
Jun 12
spent yesterday at @Cloudflare Connect on Tour in Vancouver. a few things that stuck with me. thread 🧵
1
1
245
Jun 12
the stat of the day: devs now controls ~100 autonomous agents (average across a few internal and external speakers). one compromised dev account is no longer one compromised dev account. blast radius problem nobody has solved.
1
1
124
Jun 12
Surprised to see that only 50% of internet traffic is now automated. the old "human good, bot bad" binary is dead. @Cloudflare scores every request 1-99 on behavioral trust instead.
1
207
Jun 12
Really enjoying Fable for nuanced tasks technical content review. knowing when to pivot to the model is key given the usage caps / costs though. How are you managing model pivots based on task?
37
Jun 10
It was only a matter of time.
NEW: malware developers added nuclear & biological weapons text to to their spyware. Goal? To trigger LLM safety refusals... so that their spyware wouldn't be analyzed by an AI security scanner. Cleanest practical example I can think of for why over-indexing on first order safety alignment is risky. When closed (and open) models ship with aggressive refusals, they will be sprinkled with second-order blindspots that attackers will discover...and exploit. We are only in the earliest days of attackers leveraging these features, and it wouldn't surprise me if users systems that need to handle complex cybersecurity issues demand that models be less safety-blunted. In the weeds: @SocketSecurity's post also shows why intention matters in how you design a malware analysis pipeline to avoid prompt manipulation. H/T to colleagues that shared this with me socket.dev/blog/mini-shai-hu…
1
99
Jake retweeted
I've got an agent in a loop optimizing a renderer with the goal to minimize frame times (and tests to measure). It got times down from 88ms to 2ms and allocations down from ~150K to 500. Sounds good, right? Wrong. This is exactly why agent psychosis is a big fucking problem. As an experiment, I rewrote the Ghostty core render state in Go, with access to identically laid out data structures as Ghostty and the exact same validation tests. I made a purposely naive renderer (simple, correct, but slow). 88ms per frame with 150,000 allocations (horrendous, lol)! I then kickstarted a Ralph loop to bring the frame times down. I told it it can't modify input data structures or the public API or tests (they're correct), but it can do anything else it wants. It got to work. It has worked for about 4 hours. I've spent around $350 on this experiment so far. The results? 88ms => 1.5ms 150K allocs => ~500 allocs Incredible right? Nope. My hand-written renderer I ported has frame times (same benchmark) of ~20us (0.020ms) and 0 allocations in the update path. This is the problem with psychosis and lacking systems understanding. If you don't understand the system, you're going to accept that this is an incredible result. If you understand the system, you'll see better solutions immediately and can do roughly 75x better on throughput. The people who blindly trust agent output are in the former camp. They're sheeple, overdrinking from a fountain of mediocrity. Standard disclaimer: I use AI all the time. I like AI. The point I'm making is to not blindly accept results. Think. Analyze. Learn.
308
979
8,937
791,102
Jun 2
some great talks at @BSidesVancouver this week. kudos to the crew / organizers. killed it with some really good topics!
3
146
May 26
press c to copy on the tui is golden. press c to copy from the browser would be nice too @AnthropicAI ?
1
3
202
May 26
Is Toronto Tech Week a must attend event in 🇨🇦 these days? Its all over my feed.
2
3
523
May 26
🇨🇦Toronto Tech Week checkin🇨🇦 whats the best thing you've found today at tech week Drop a link 👇
1
14
2,872
May 26
context window on 4.7 is unreal.
2
3
784
Jake retweeted
May 24
canadian bootstrap founders, i’m putting together a list of cool canadian bootstrap founders/products i know @chatbase by @yasser_elsaid_ , @postbridge_ by @jackfriks , @usecollabstr by @dulay604, @obsdmd by @kepano, @indexsy, and @yacineMTB stuff are popular ones any other canadian bootstrap companies/founders i should know about?
68
6
158
12,039
May 22
ran out of tokens, went for a ride. @AnthropicAI is out here caring for your wellbeing by adding limits.
4
139
Jake retweeted
Thank you to @IMDAsg and @AsiaTechxSG for hosting us in Singapore this week! The skeptics who called AI a bubble last year saw breakthroughs this year - can't wait to see where we'll all be in 2027!
7
3
19
888