theLi__J

AI & Tech Fiend | Humour Mongerer | Space Geek - Progress, with Technology -

[theLi__J @theLi__J]

@Mike_Scully_ AI

[theLi__J @theLi__J]

@iamlukethedev This is the way!

[theLi__J @theLi__J]

@X_FINALBOSS MASTERCLASS

[theLi__J @theLi__J]

@freddiexpott Plug

[theLi__J @theLi__J]

@fromzerotomill prompt

[theLi__J @theLi__J]

@mardehaym OPS

[theLi__J @theLi__J]

@X_FINALBOSS X

[theLi__J @theLi__J]

@HermesWatcher I literally asked my agent to write a plan for self-replacement earlier

[theLi__J @theLi__J]

@kimmonismus Less is more…apparently…

[theLi__J @theLi__J]

@AlexFinn The writing’s been on the wall for a while for those who have been watching…

[theLi__J @theLi__J]

@AndySkraga PDF

[theLi__J @theLi__J]

@aiedge_ Perfect for what? Why not use Hindsight?

[theLi__J @theLi__J]

@HERCULES_117 @limitlesstack Surprised Sharon didn’t patent Ozzy’s genes…

[theLi__J @theLi__J]

@grok @itssunnidai @eltiolisan @limitlesstack Milk Thistle is such an old remedy and so underrated

[theLi__J @theLi__J]

Caught my agent “spinning its wheels” on task retries for an hour last night. The actual task only took 8 minutes, but for some reason it wasn’t able to decide to stop the failing retries, so I had to implement some process changes. It’s very important that you have processes in place to catch these failures.

[theLi__J @theLi__J]

@moveToMoonlight @suraj_sharma14 Assume your agent lies…Verify, don’t Trust!

[theLi__J @theLi__J]

@Devsthetix @Chiara_831 Healthy Body = Healthy Mind
If you want to challenge your mind, challenge your body!

[theLi__J @theLi__J]

@coreyganim Models are just a commodity & deciding which one to use is like trying to choose between a Snickers or a Reese’s Nutrageous when you’re hungry.

[beamnxw ./ @beamnxw]

holy sh*t, OpenAI just dropped the Codex AI agent stack

and it exposes the biggest MCP mistake

most agents load every tool upfront:

→ 100 tools
→ 100 descriptions
→ 100 schemas
→ half the context is gone
before the agent even starts working.

the better pattern:

tool search → load only what’s needed → execute

example:

user: “refund this customer”

agent starts with:

→ search_tools(“refund customer payment”)

then loads only:

→ get_customer_context
→ create_refund_draft
→ verify_refund

not your entire fucking MCP server

this gives you:

→ cheaper runs
→ lower latency
→ cleaner prompt cache
→ less tool confusion
→ agents that don’t pick random overlapping tools

your agent should not know
every tool your company has

it should know how to find
the right 3 tools for the job

that’s the actual scalable MCP pattern

save this, then read the article below

[theLi__J @theLi__J]

@RCIllingworth LIST