Firetiger (@usefiretiger) / X

Firetiger

26 posts

Firetiger

@usefiretiger

Observability is dead. Long live outcome engineering. You and your AI agents write code. Firetiger makes sure it works.

San Francisco

Joined August 2024

Firetiger reposted
Rustam X. Lalkaka
@lalkaka
Jun 8
driving continuous improvement loops in software systems is really hard, even if you have really smart agents (codex, claude, cursor), really smart humans, decent observability tools, etc. today we're launching Service Monitors, loops in prod made easy. blog.firetiger.com/firetiger-serv…
Matt Van Horn
@mvanhorn
Jun 8
Article
WTF Is a Loop? Peter Steinberger vs. Boris Cherny
The most repeated sentence in AI coding this week is six words long, and almost nobody saying it can define it. One tweet had the entire timeline in a chokehold this week, so I ran /last30days on the...
1.6M
Firetiger
@usefiretiger
Jun 2
Nobody actually cares about slow queries. They care that checkout is broken. Firetiger agents start from the customer impact, then trace issues back to the technical root cause. #OutcomeEngineering #AgenticAI #DevOps
00:00
104
Firetiger
@usefiretiger
May 28
Shipping code faster is easier than running it safely in production. @usekernel is building infrastructure for AI agents, which means things move quickly, and reliability matters a lot. In this post, the Kernel team explains how they use Firetiger to monitor deployments, catch
How Kernel keeps shipping fast with Firetiger
From blog.firetiger.com
127
Firetiger reposted
Rustam X. Lalkaka
@lalkaka
May 19
effective LLM prompt caching makes or breaks inference budgets. tuning cache TTLs is hard and fiddly. we pointed persistent background agents at the task and saw a 77% reduction in cache write waste in a week. more on how here:
Agentically optimizing LLM prompt cache TTLs for fun and profit
From blog.firetiger.com
743
Firetiger reposted
Achille
@__Achille__
May 14
@TownAI only had to connect @AxiomFM and @convex to their @usefiretiger agents to get high-quality change monitors setup. Now an agent joins each of their PRs, drafting a monitoring plan for what to lookout for when changes get deployed, and reporting with RCA on every issue.
205
Firetiger
@usefiretiger
May 12
“We pay close attention to every Firetiger notifications.” 🫳🎤
Achille
@__Achille__
May 12
It's been amazing hearing from engineers like @georgekontridze from @TownAI about how they've adopted @usefiretiger
00:00
171
Firetiger
@usefiretiger
Apr 27
Everyone is using AI to write code now. So what breaks next? - Code review - Trust - Observability In EP02 of Canned Responses, @lalkaka invited @templaedhel (CTO @ Metronome / Stripe) to taste tinned fish and talk about the future of engineering.
00:00
400
Firetiger
@usefiretiger
Apr 27
Watch the full episode here:
110
Firetiger reposted
Achille
@__Achille__
Apr 17
Replying to @__Achille__
Object store as a primitive for Agent runtime
From blog.firetiger.com
178
Firetiger reposted
Achille
@__Achille__
Apr 10
I heard in several conversations that coding agents are the end of open-source software. I strongly disagree.
389
Firetiger reposted
Rustam X. Lalkaka
@lalkaka
Apr 7
a fun new feature! "fun", perhaps. defining and tracking Service Level Objectives are never fun! UNTIL TODAY! @usefiretiger agents now identify good service level indicators and set objectives intelligently when you set them forth on a mission to improve an outcome in
227
Firetiger reposted
Achille
@__Achille__
Apr 1
A few weeks ago, @lalkaka threw this a new idea out: “Let’s interview tech leaders… while eating canned fish.” You think this was a joke, it isn’t, and everyone we talked to about it immediately loved the idea! So we filmed it and it’s called Canned Responses. Episode 01 is
00:00
953
Firetiger reposted
Rustam X. Lalkaka
@lalkaka
Mar 31
we run into the same compaction bug and have the same hack in our agent harness @usefiretiger . fun to see the emperor has the same hole in their clothes.
alex000kim.com
The Claude Code Source Leak: fake tools, frustration regexes, undercover mode, and more
Anthropic accidentally shipped a source map in their npm package, exposing the full Claude Code source. Here's what I found inside.
396
Firetiger
@usefiretiger
Mar 30
Large tool outputs quietly kill agent performance. You call an API, pass everything to the model, and hope for a clean answer. You get timeouts or messy reasoning instead. @swnelson_ from Firetiger breaks down how we handle it: - Truncation, which appears simple at first (but
Agent Engineering Patterns: Dealing with very large tool results
From blog.firetiger.com
452