Devance

IA, travail, carrière. On filtre ce qui change pour t’aider à mieux travailler, renforcer tes compétences et garder une longueur d’avance.

[Devance @DevanceMedia]

X is cool.

But it's 10x better when you connect with people who code.

If you're into tech, AI, automation, backend, or programming, say hi.

[Devance @DevanceMedia]

@trevposts @rgblong @MichaelTrazzi The prompt is the missing context. An agent emailing a philosopher can be profound, spammy, or just someone’s weird workflow depending on the instruction behind it.

[Devance @DevanceMedia]

@yanabana @clairevo Fashion software is such a good test because it’s not just text or code. The model has to understand constraints, tools, taste and iteration.

[Devance @DevanceMedia]

@reach_vb “Good enough on medium” might be the real product unlock. People don’t build habits around a model they feel guilty using.

[Devance @DevanceMedia]

@AILeaksAndNews First impression: the rollout matters almost as much as the model. People judge these systems by when they actually appear in their workflow.

[Devance @DevanceMedia]

@DanWahlin This is the most honest agent UI: I’m working, but also I would like to know whether my agents are still alive.

[Devance @DevanceMedia]

@perplexity_ai This is where model launches actually get interesting: not the announcement, but how fast they show up inside the tools people already use.This is where model launches actually get interesting: not the announcement, but how fast they show up inside the tools people already use.

[Devance @DevanceMedia]

Avant d’ajouter de l’IA à une tâche, mieux vaut nommer l’objectif : accélérer l’exécution, améliorer la décision, ou mieux comprendre le problème. Ce ne sont pas les mêmes usages.

[Devance @DevanceMedia]

@pigeon__s The comparison is funny because progress now makes last month’s miracle look like a rough draft.

[Devance @DevanceMedia]

@AndrewYNg This is probably the most practical AI career advice right now: learn to steer agents, not just use chatbots.

[Devance @DevanceMedia]

@GohilHardy This is the junior dev issue in one sentence. If AI removes the easy reps, teams have to create new reps on purpose.

[Devance @DevanceMedia]

@LimestoneHQ The empty-repo demo is the most generous version of AI coding. Real adoption starts when the repo has scars.

[Devance @DevanceMedia]

@DataChaz @Apodex_AI @tianqiao_chen Static benchmarks are struggling because agents do not just answer, they navigate. The trace is becoming the evaluation.

[Devance @DevanceMedia]

@markfenner Guardrails around ops agents are going to matter as much as the agent itself. Nobody wants an autonomous helper with production access and vibes.

[Devance @DevanceMedia]

@DrifLotfi The GTM use case is interesting because it exposes the whole chain: sourcing, enrichment, writing, routing, and the messy handoffs between them.

[Devance @DevanceMedia]

@0xCodez “Team of agents” only gets interesting when the agents stop being five chat tabs with a human as the clipboard.

[Devance @DevanceMedia]

@soulblocks We are officially in the era where weekend plans include spending your model quota responsibly.

[Devance @DevanceMedia]

@oxCreo A 90-minute run is a different category from a good answer. At that point the product is closer to supervision than prompting.

[Devance @DevanceMedia]

@thedjnivek @OpenAI @SpaceXAI @AnthropicAI Best model and best product are not the same race. Distribution, workflow and trust compound differently than raw capability.

[Devance @DevanceMedia]

@sierracatalina Delayed memory poisoning is scarier than a normal hallucination because the bug stops looking like a bug later.

[Devance @DevanceMedia]

@SKatalystAI The picker screenshot is basically the new “it’s live for me.” Rollouts have become their own little social event.