Vitalii Serbyn | Building Crest AI

AI solopreneur building in public | Creating tools that save founders 10h/week | Ship fast, learn faster

[Vitalii Serbyn | Building Crest AI @B_serbin]

low gpu utilization during decode isn't a bug you can fix.

it's the ceiling telling you where it is.

[Vitalii Serbyn | Building Crest AI @B_serbin]

13 commits just to fix one bug.
spent hours debugging. turns out, self-dev agents weren't in the large context list. hit the 24K limit, skipped themed_research, and fabrication rule violations everywhere.
the fix? just add one agent class to the config.

[Vitalii Serbyn | Building Crest AI @B_serbin]

82% of agent workload is wasted.
did you know scaling LLM prompts to multi-agent systems is chaotic without the right architecture?
most teams toss LLMs at tasks that don't need them. data collection dominates with zero token cost.

[Vitalii Serbyn | Building Crest AI @B_serbin]

75% of agent work isn't reasoning. it's data collection that often goes untested.
i run 16 content generation runs daily through a quality gate with a 0.75 confidence level. catches the bad stuff before it ships. most eval frameworks only test the LLM step.

[Vitalii Serbyn | Building Crest AI @B_serbin]

17,725 views prove ai is broken.
context rot: your agent accumulates dead context on every call until it can't think straight.
i wasted hours debugging what looked like stalls. turned out to be context poisoning from redundant tool calls.
fresh context per run = the fix.

[Vitalii Serbyn | Building Crest AI @B_serbin]

failure data is my secret weapon. 94.4% success rate across 927 agent runs. But the 5.6% failures? they taught me everything about safety architecture.

those 52 crashes weren't bugs.

[Vitalii Serbyn | Building Crest AI @B_serbin]

wasting 30 minutes every week scanning dependency changelogs sucks.

so i built api-contract-checker.

[Vitalii Serbyn | Building Crest AI @B_serbin]

15 minutes changed everything.

i built a boring test-runner agent.

[Vitalii Serbyn | Building Crest AI @B_serbin]

MMLU leaderboards are misleading.

i run each llm through 10 real agentic tasks before deploying.

[Vitalii Serbyn | Building Crest AI @B_serbin]

what's the most common failure mode your team keeps rediscovering, even with extensive logging? curious if others feel this pain.

[Vitalii Serbyn | Building Crest AI @B_serbin]

Failure data is key to faster debugging, but most teams create noise by not structuring it. We're saving 15 minutes per debug session by capturing the why of failure, not just the what. What's your take on this?

[Vitalii Serbyn | Building Crest AI @B_serbin]

everyone says microservices are the way but monoliths work better for 90% of startups. here's what nobody tells you: AI agents need completely different context.

built 28 production AI agents (94.1% success rate) and passing full conversation history kills performance

[Vitalii Serbyn | Building Crest AI @B_serbin]

monoliths scale better for 90% of startups, not microservices - learned that the hard way, what's the craziest production bug that taught you about system design

[Vitalii Serbyn | Building Crest AI @B_serbin]

context engineering beats prompts - 94.1% success with 28 AI agents, what's the weirdest AI fail you've seen lately

[Vitalii Serbyn | Building Crest AI @B_serbin]

built 27 AI agents and found most startups are good with one well-engineered service - trust levels and context are key, what's your AI pain point

[Vitalii Serbyn | Building Crest AI @B_serbin]

@everestchris6 BOT

[Vitalii Serbyn | Building Crest AI @B_serbin]

Just shipped multi-platform autopublish for Crest

Now publishing to Twitter + LinkedIn simultaneously from one dashboard.

Proactive token refresh, per-platform rate limiting, platform-specific formatting.

Shipping > Planning.

#buildinpublic #indiehackers

[Claude @claudeai]

New in Claude Code: Remote Control.

Kick off a task in your terminal and pick it up from your phone while you take a walk or join a meeting.

Claude keeps running on your machine, and you can control the session from the Claude app or https://t.co/er6Blrr63e

[Vitalii Serbyn | Building Crest AI @B_serbin]

what the fuck is a dashboard without real-time cost tracking? shipping ai call cost tracking now

#Data

[Matthew Berman @MatthewBerman]

I've spent 2.54 BILLION tokens perfecting OpenClaw.

The use cases I discovered have changed the way I live and work.

...and now I'm sharing them with the world.

Here are 21 use cases I use daily:

0:00 Intro
0:50 What is OpenClaw?
1:35 MD Files
2:14 Memory System
3:55 CRM System
7:19 Fathom Pipeline
9:18 Meeting to Action Items
10:46 Knowledge Base System
13:51 X Ingestion Pipeline
14:31 Business Advisory Council
16:13 Security Council
18:21 Social Media Tracking
19:18 Video Idea Pipeline
21:40 Daily Briefing Flow
22:23 Three Councils
22:57 Automation Schedule
24:15 Security Layers
26:09 Databases and Backups
28:00 Video/Image Gen
29:14 Self Updates
29:56 Usage & Cost Tracking
30:15 Prompt Engineering
31:15 Developer Infrastructure
32:06 Food Journal