Steve | Systems & AI
Region: US
Friday 20 February 2026 14:27:23 GMT
4071
176
19
11
Music
Download
Comments
SoundsOfSepehr :
Real question is how are you affording tokens for these agentic workflows. I used Sonnet to manage my project with OpenClaw and went through $20 in one day.
2026-02-20 20:34:06
0
sonicpoet :
Not knowing bothers me. Thats what got me into this business in the first place. Automation is one thing. I don’t trust the agents. Been burned too many times in the past. The mishaps are less common than they used to be, which is arguably worse.
2026-02-21 02:39:34
1
Jay Long :
I agree but I'm noticing a trend where reasoning tokens are being hidden by default but still available if you expand for details.
2026-02-21 04:28:10
0
nycmike12300000 :
make mcp pub/sub inject logging use MQ or Kafka .. can even use connectors to log too
2026-02-20 22:28:44
0
Cato :
This is what I couldn’t put in words. Watching AI work actually teaches you a lot of things you don’t know. In the end you’re even capable of steering it better because you’re in sync.
2026-02-20 17:38:00
1
13K_ :
And that’s were as a software engineer you come in to fix. Instead of overviewing what it does. Why can you fix it after it’s done everything. Think of it like a company giving you a system to maintain it but you never worked on it before
2026-02-20 19:14:34
0
TechWitchTitan🏳️🌈👽🛠🖖🏖⛈️ :
I built this to solve the language barriers between human and machine https://www.npmjs.com/package/project-arch
2026-03-15 13:49:51
0
specvsnt :
This is such an excellent and clear explanation of how I feel about my current multi-agent Claude workflow. I couldn’t put it into words as clearly when speaking with team members, but you hit the nail on the head. Excellent videos!
2026-02-20 18:28:39
1
Hint :
You are on point
2026-02-20 19:21:42
1
same_trapz :
💯
2026-02-23 17:47:57
0
monatemediaofficial :
At last someone who AI and is honest. Later, when you do work on the system, you’re not as confident about your changes and the effect it will have on the system. You also find that even though your syntax or reasoning may be inefficient if you hand coded it, at least you understand what you did, where this thing is so efficient, it just omits syntax that makes the code understandable. Then you throw it in the trash can and build a new version, this time, you use a framework.
2026-02-20 14:42:50
1
Aljo :
to every single agent sys I built I always add tree of thoughts to be displayed, decisions, entire thought process, tool selection and so on. witouth it it would be pretty much guessing what's going on and how agents you build work actually. recently I forced it to display even why task dcomposition is happening the way it happens, prompt extraction in massive prompts and even why and how he planed those tasks. it's pretty interesting to see it and actually tels me what is needed to do or enhance or adjust. imagine making agent blindly not knowing what it does. omg. most of times it's huge issue with prompt extraction that fails rest of process or Gemini models with parsing and failing to use tools 😂
2026-02-20 22:07:50
1
sitetrust :
Spot on 👌🏼👌🏼
2026-03-31 19:42:32
0
To see more videos from user @sd141414, please go to the Tikwm
homepage.
