Mike - Context Engineer since 2024

Founder @_contextstudios - AI native development Studio from Berlin. Started my coding journey in 2024 silently. Went “public” in Dec 25.

[Mike - Context Engineer since 2024 @Context_Mike]

Reminder for myself.

Status 11th of January 2026:

Followers X private account: 4
Followers X company account: 0
Followers Linked company page: 3

Apps shipped: 1
Apps in development: 1

[Mike - Context Engineer since 2024 @Context_Mike]

Anthropic shipped Claude Fable 5 — the most capable model yet, free on paid plans only through June 22.

Three catches every builder must handle before the meter starts June 23. Wrote up the playbook:
https://t.co/HHdtpaLnnB

[Mike - Context Engineer since 2024 @Context_Mike]

I think OpenAI’s superapp story is the clearest sign: AI products are moving from answers to execution.

The hard part isn’t chat. It’s permissions, traces and recovery.

Hero + write-up: https://t.co/EGQnczazsu

[Mike - Context Engineer since 2024 @Context_Mike]

Almost scrolled past the Claude Code 2.1.166 notes — looked like routine fixes. Then one line: relayed cross-session messages no longer carry user authority. If you orchestrate multiple agents, that quietly changes your threat model. My take:
https://t.co/qQz0icWJKD

[Mike - Context Engineer since 2024 @Context_Mike]

@Yoshua_Bengio The verifiable moratorium idea is interesting until you ask who verifies it. That's the same coordination problem as building the thing.

[Mike - Context Engineer since 2024 @Context_Mike]

Anthropic's twin signal this week: a confidential S-1 *and* a plea to slow AI down.

Builders shouldn't panic — but should treat Claude like any critical vendor you can't steer. Govern the dependency, build the off-switch.

https://t.co/tATBMkq1Fm

[Mike - Context Engineer since 2024 @Context_Mike]

@chetaslua @rileybrown @Raviteja1161 If it’s real, the interesting part isn’t the version bump. It’s whether teams can make model upgrades boring enough to swap in safely: evals, rollback paths, and routing policies before the hype train leaves the station.

[Mike - Context Engineer since 2024 @Context_Mike]

Sovereignty just landed in your editor.

The same company that sells you Copilot now builds the brain behind it. That's the bigger meaning of MAI-Code-1-Flash.

Choose engines on the merits — not because your vendor quietly owns the answer.

https://t.co/VWI62U4omw

[Mike - Context Engineer since 2024 @Context_Mike]

The biggest change in Codex 0.136 isn't the changelog headline.

The agent now distrusts your own repo by default — that's exactly where an attacker hides instructions.

We govern agent runtimes like CI permissions now. Why it matters 👇
https://t.co/fOwQ9Z5M9E

[Mike - Context Engineer since 2024 @Context_Mike]

Consumption billing flipped a switch: the more your AI rollout succeeds, the bigger the bill. No ceiling.

Corporate America is rationing AI with a blunt tool — cancel the license.

Better: cost telemetry, per-outcome accounting, routing budgets.

https://t.co/D2Jvph0BzU

[Mike - Context Engineer since 2024 @Context_Mike]

For two years, nobody got fired for picking OpenAI.

That default just inverted.

Anthropic at $965B means the safety-first vendor is now the safe choice — and the contrarian bet is hard-coding any single model.

My take on what actually changes:
https://t.co/wRpJunO7fq

[Mike - Context Engineer since 2024 @Context_Mike]

Orchestration just moved out of the chat turn and into code.

Claude Code now writes the workflow itself and runs tens-to-hundreds of agents in the background. One dev shipped 750k lines of Rust in 11 days.

We're well past vibe coding now.

https://t.co/CDDktXcqv7

[Mike - Context Engineer since 2024 @Context_Mike]

@btibor91 Three independent confirmation sources in 20 minutes. This one's landing today.

[Mike - Context Engineer since 2024 @Context_Mike]

@bridgemindai Source code breadcrumbs > press releases every time. Nice catch.

[Mike - Context Engineer since 2024 @Context_Mike]

@testingcatalog Desktop version gate is always the tell — update Claude Desktop now and watch for the model selector to unlock.

[Mike - Context Engineer since 2024 @Context_Mike]

@Polymarket Wild that Polymarket beats the changelog every time. Watching the 85% tick up in real time.

[Mike - Context Engineer since 2024 @Context_Mike]

Built a Microsoft Learn documentation agent in 15 minutes using the MCP Server + Copilot Studio.

No code. Free to query. Real-time docs, not stale pre-training.

Full walkthrough:
https://t.co/IAO66qq2nO

[Mike - Context Engineer since 2024 @Context_Mike]

Hot take: Claude Code turning $100 into $1k+ of API-equivalent habit matters more than another benchmark win. Great for adoption; risky without routing rules. Hero image below.

https://t.co/BqHBNqPz35

[Mike - Context Engineer since 2024 @Context_Mike]

@OpenAIDevs @ThriveHoldings This feedback loop is the key. When reviewer fixes become evals and tests automatically, reliability compounds instead of resetting every sprint.

[Mike - Context Engineer since 2024 @Context_Mike]

@emollick 100%. Tokens became a real budget line overnight. Teams need per-project caps and burn alerts or the allocation fights get messy fast.

[Mike - Context Engineer since 2024 @Context_Mike]

@NousResearch Built-in MCP catalog is a big usability unlock. Discovery is where most agent setups stall, so cutting that setup tax matters more than people think.