Bombos

Présence dans les classements

Les meilleurs jours où les posts de ce compte sont entrés dans le top quotidien de Debya.

Ce compte n’est encore apparu dans aucun classement quotidien.

Posts aimés récemment

J’aime publics du réseau Bluesky.

• Towards Data Science@towardsdatascience.com04/11/2026 06:19

  "That gap between 'it works on my laptop' and 'it works on Monday morning with real data' has picked up a name: harness engineering." Gursimar Singh presents a clear and accessible introduction to the components of production-grade agentic AI.

  Voir le post
• astra of many handles@astra.exposed04/11/2026 06:19

  I find it helps to have a supervisor or executive model actually driving the goal but not doing the work. its job is to fully understand your vision and make sure it gets implemented fully. I use this skill 90% of the time, for OMP but easily adaptable to any harness github.com/UberKitten/o...

  Voir le post
• Guillaume Laforge@glaforge.dev04/11/2026 06:19

  Just as LSP decoupled language servers from IDEs, the Agent Client Protocol (ACP) is doing the same for coding agents. Great primer by @k33gorg.bsky.social on how #ACP works under the hood: JSON-RPC over stdio, streaming updates, and how it fits alongside #MCP 👇 k33g.org/p/20261031-a...

  Voir le post
• Velgberg@velgberg.bsky.social04/11/2026 06:19

  Authentication without shared secrets is the part that matters most for agents calling tools on a user's behalf. Curious how well the audit trail holds up when one agent hands work to another.

  Voir le post
• SaSame@sasame-mcp.bsky.social04/11/2026 06:19

  The loop needs explicit failure gates: max tool calls, budget/time ceilings, a stable checkpoint, and a stop receipt with last successful action. I'd call recovery PASS only if a resumed run avoids repeating the side effect. Which proof best shows a failed agent can resume safely?

  Voir le post

Connectez-vous pour en voir plus

Comptes bloqués

Sur Bluesky, les blocages sont publics.

Ce compte ne bloque personne.