Akash

I give agents real work and see what breaks.
Building | SDE

[Akash @akashcorex]

What happens when AI starts doing the work?
Like really replacing something

That's what I'm trying to figure out. It's like finding One piece in real life.

I build agents, test them on real-world problems, and experiment with new models, tools, and workflows.

Sharing what works, what breaks, and what I learn along the way. Only raw, no bluff

Less hype, more practical.

[Akash @akashcorex]

@thorstenball m'agent is the only polite greeting that still feels like a threat

[Akash @akashcorex]

@Hesamation has to work on domestic chips is the real roadmap. parameter count is the press release. supply chain is the product

[Akash @akashcorex]

@jasonlk carefully is doing all the work in that sentence. most sdr demos skip the pricing + objection bank and then act shocked when outbound sounds like a brochure

[Akash @akashcorex]

@FlorinPop17 curious which change actually broke ur muscle memory. for me its always the sidebar that moved 8px and ruined muscle memory

[Akash @akashcorex]

@shydev69 connectors that paint a jpeg of an empty calendar instead of canceling meetings. the vision model worked. the agent never left chat

[Akash @akashcorex]

open weight demand is real. it just isnt the same chart as frontier wallets. middlemen got a second customer. the labs that eat mistakes still get the margin

[dax @thdxr]

people are struggling to understand that "token middlemen have seen growth primarily from open weight model demand" isn't the same thing as "frontier token demand is dying"

[Akash @akashcorex]

@OmriBuilds quota theater wins again. model quality barely got a hearing

[Akash @akashcorex]

@damonchen customers buy outcomes not model brands. jev is ur problem not theirs

[Akash @akashcorex]

@antirez hype picks the smallest shiny thing and calls it the milestone. jev is a tool. the bubble treats every tool like a paradigm shift

[Akash @akashcorex]

@GergelyOrosz day 1 dopamine then radio silence is the most honest agent eval ive seen. the model worked. ur decision avoidance just got a mirror

[Akash @akashcorex]

@vineerpasam ai code cleaner banned from using ai is the only role that still needs taste. hiring that person harder than hiring the one who made the slop

[Akash @akashcorex]

@OmidTheTechGuy pretty chat is the trap. once approvers need policy codes + evidence trail, the ui stops being a toy and starts being the product

[Akash @akashcorex]

@vintcessun deal signed. comedy is now a stakeholder with write access to the acceptance criteria

[Akash @akashcorex]

@Saisharan0099 exactly. silence trains cargo-cult keywording. a coarse reject reason is the cheapest eval loop hiring never ships

[Akash @akashcorex]

every job app is a new resume rewrite
same person. same skills. different keywords so an ats can pretend it read u

coding agents get evals overnight. job apps get silence and a new version of the same bullet

who owns the feedback loop when the only signal is a ghost?

[Akash @akashcorex]

@Trade_at_Arrow Most cheap ass add i ever saw

[Akash @akashcorex]

@fdotinc 0.0057 cpt is cute until the store changes one selector and ur cost chart becomes a support queue

[Akash @akashcorex]

@omarsar0 interactive primer > another thread explaining jev. playground is the right format

[Akash @akashcorex]

@yongfook the moat was never time. it was owning the boring glue forever. agents shrink the first half and leave the glue tax untouched

[Akash @akashcorex]

@rauchg ephemeral vercel deploy into an iphone sim just to prove a pixel lied. humans wouldve shrugged and said works on my machine by tweet 2

[Akash @akashcorex]

@addyosmani curious if those eval cases catch real regressions or just score the same demo prompt the plugin was tuned on