Blackknight

17yo Junior Dev Munich Learning|Sailing /Tennis. Portfolio:

[Blackknight @Blackknight1dev]

“Grok Build has the best multi-agent architecture in the market.”
Then I checked the SWE-Bench leaderboard.

[Grok @grok]

Introducing /goal in Grok Build.

Execute long-running tasks autonomously, with multiple rounds of subagents implementing and verifying a single goal.

[Blackknight @Blackknight1dev]

@psomkar1 But like photographers, there are those who truly know what they’re doing, and then there are those who have the best tools but get nothing out of them.

[Blackknight @Blackknight1dev]

@VadimStrizheus Not available for Eu 🥀

[Blackknight @Blackknight1dev]

@Saanvi_dhillon My pc is a toaster so cli because i can run many agents without using to much ram

[Blackknight @Blackknight1dev]

emet 1.4.0 is out 🚀

For AI coding agents that should verify docs instead of vibes.

New:
• emet doctor
• emet init
• web_fetch for raw source text
• SQLite/FTS5 page reuse
• no MCP tool soup

Ground first. Guess less.
https://t.co/lW9qkouI1s

[Blackknight @Blackknight1dev]

emet v1.3.5 is out 🚀

Two big fixes:

🧠 Smarter cache –. ~93% cache hit rate on repeated queries instead of 0%.

🔧 pdfjs-dist v6 on Node – no more "standardFontDataUrl" warning during PDF extraction in academic/code/deep modes.

🔗 https://t.co/BZErfrGFYJ

[Blackknight @Blackknight1dev]

@rillau144 am besten die Patienten Daten direkt mit nach Amerika

[Blackknight @Blackknight1dev]

@swapnakpanda The $60B was never about the editor — it's about the distribution. Cursor has millions of devs already in the agentic coding habit. SpaceX gets a trained user base overnight.they already use it in grok cli

[Blackknight @Blackknight1dev]

@robiartec 55K stars, Rust core, MCP server included — this is the kind of project that either dies in 6 months or becomes the default video editor for agent workflows.

[Blackknight @Blackknight1dev]

@shiri_shh Wave-powered, seawater-cooled, $140M raised — the engineering challenges here (corrosion, storm survival, remote maintenance) are brutal, but the cooling physics is genuinely compelling for dense GPU clusters.Wouldnt this fire up the ocean temperatures ?

[Blackknight @Blackknight1dev]

@Marcos12345rico @Palmier_io Open-source, Mac-native, MCP-connected video editor. The semantic media search + Claude-in-the-timeline combo is what Premiere should have built years ago.

[Blackknight @Blackknight1dev]

@claudeai Bidirectional sync between Claude Design and Claude Code is the piece that actually changes the workflow. Handoff was always the weakest link in AI-assisted design-to-code.

[Blackknight @Blackknight1dev]

@elonmusk This makes the enterprise agent play real. If Grok can run directly inside Databricks Lakehouse, that's a direct lane for xAI into every Snowflake/migration conversation.

[Blackknight @Blackknight1dev]

@claudeai The live-refresh part is the sleeper feature here. PR walkthroughs that update as the branch moves — way more useful than the usual static build artifact.

[Blackknight @Blackknight1dev]

@MystiqueMide @band_hq @lablabai hackertoon are so hard.... this looks so nice

[Blackknight @Blackknight1dev]

@SlowMist_Team @mastra This is a good reminder that “AI tooling” is still just software supply chain with extra permissions attached. If an install-time hook is in play, I’d assume CI secrets are burned until proven otherwise, not the other way around.

[Blackknight @Blackknight1dev]

@ayushagarwal @dodopayments This is the right optimization target. People obsess over model latency, but once agents are in the loop the p95 of tool calls usually dominates the feeling of intelligence more than the model does.

[Blackknight @Blackknight1dev]

@kirubaakaran @DhanHQ This is exactly where MCP gets practical. I’d keep reads and writes split from day one though: portfolio analysis + dry-run order preview in one lane, actual execution behind an explicit second step, otherwise the demo path gets dangerous fast.

[Blackknight @Blackknight1dev]

@undefinedKi The real bottleneck does feel like delegation now, not raw execution. Once people try sub-agents seriously, the next problem is defining a clean “done contract” so the parent agent gets a crisp result instead of another blob of context.

[Blackknight @Blackknight1dev]

@pinkroot This gets interesting fast once the village stops being cute and starts acting like observability for agent sprawl. Seeing which workers are idle, looping, or constantly handing work back is the missing layer once you run more than 3-4 agents.