Mika

Coding with AI. Learning, building, and sharing the process.

[Mika @mika_systems]

https://t.co/EvoaFa63Cd

[Mika @mika_systems]

@EugBass thanks

If your comment receives a lot of likes, I will continue to publish similar documents

[Mika @mika_systems]

we've been paying AI to read the same document 13 times

TypeSafe gave Jev a 53,777-character GDPR article and 13 questions: 8 Noul, 2 Choice and 3 Score.

they tested two methods, five times each:

> 13 separate calls, one question each
> 1 Jev call containing all 13 questions

batching didn't measurably change the answers. it changed the bill.

13 calls → $0.006090 → 2.71s
1 batch → $0.000497 → 0.27s

12.2× cheaper. 10.0× faster than sequential calls.

the reason is almost stupidly simple: each separate call resends the entire article. Jev can read it once and score all 13 questions against the same state.

if you're building this into an agent:

> put the shared context into one state
> define independent decisions up front
> batch Choice, Score and Noul questions
> execute high-confidence decisions in code
> send uncertain cases to a human

11 of 13 tracked answers were identical across all five runs. the other two varied slightly under both methods.

if your agent asks several questions about the same document, let Jev read it once.

save the complete 4-page Jev field note - then read the full 9-step roadmap and four agent patterns in the article below ↓

[Mika @mika_systems]

https://t.co/EvoaFa63Cd

[Mika @mika_systems]

@EnergyJoltou ahaha that's actually true

[Mika @mika_systems]

someone finally fixed the expensive part

a new AI called Jev was released only a few days ago. it doesn’t generate paragraphs - it makes typed decisions that software can use directly

TypeSafe says Jev is up to 193.6× faster and 444.6× cheaper than frontier models in selected workflow tests

the listed price is $0.042 per million input tokens. output tokens are free

that changes the economics of agents. instead of spending a full LLM call on every yes-or-no branch, Jev can classify, route or score the entire batch

one author-reported example:

700 leads → match / no match → 40 seconds → $0.09

business lift is still unverified, but the decision itself is measurable and easy to audit

the setup:

> define allowed outputs
> batch the inputs
> set a confidence threshold
> execute high-confidence decisions
> send uncertain cases to a human

four patterns already map the idea:

slop detector → score or null
UI selector → choose a component
Mac intent → choose an action
lead matcher → score 700 leads

the 28-second video follows:

FIND → TYPE → BUILD → SHIP

LLMs write and explain. Jev decides. software executes

the full 9-step blueprint and all four patterns are in the article below ↓

[Mika @mika_systems]

https://t.co/EvoaFa63Cd

[Mika @mika_systems]

@Argona0x you are 100% right

[Mika @mika_systems]

@ericmaskfr no I’m the one who should be thanking you for the support

thank you Eric

[Mika @mika_systems]

@recketino which signal would tell you this is becoming a 1980 style inflation cycle?

[Mika @mika_systems]

@recketino yes that is exactly what I am trying to explain

[Mika @mika_systems]

@HeyYoGB thank you

[Mika @mika_systems]

@KirSolo193083 the agents didn’t need a rebuild, just better questions

[Mika @mika_systems]

@KirSolo193083 I am happy to share such useful information <3

[Mika @mika_systems]

@hammertime_one the video turned out really funny

I hope the robot manages to make a pizza hah

[Mika @mika_systems]

@hammertime_one the best optimizations always look obvious once someone bothers to measure them

[Mika @mika_systems]

@chenzhangreal ahaha thank you

[Mika @mika_systems]

@cryptocompikbro the savings should hold as long as the document fits within a single context window

[Mika @mika_systems]

https://t.co/4CAsJZlKX4

[Mika @mika_systems]

@0xboan maybe you shouldn't have shown that?

just keep making while it works hah

[Mika @mika_systems]

@0xTenzo seeing the bot in action is absolutely mind-blowing tbh

[palehonk @palehonk]

WHAT, YOU DON’T LIKE MAKING F***ING GREAT MONEY?

Then why haven’t you watched IBM’s lesson on designing AI teams to handle business workflows?

Just 11 minutes:

01:32 — limit access to company resources
04:15 — give the client one interface
05:50 — organize a team of agents
09:32 — isolate their tools
10:07 — track every action

Absolutely free. Save & watch if you want to earn more with AI and build something of your own.

Want to take those skills further? My guide below walks you through building a working tool with Claude Code + Jev ↓

[Mika @mika_systems]

@egeunluu you are 100% right