@america Interesting, The CAPITALISTS and REPUBLICANS have been in power in Texas for 3 decades and control over ALL THREE FEDERAL BRANCHES for the past 2 years. Every Accusation is a Confession. All the people do is harm us and make life harder
@oliver__ai@thsottiaux@rileybrown My laptop broke, i can't use remote. And codex saves everything locally so my chat histories will be lost. With claude it's all on the cloud and I can still use cc cloud.
@realmcore_ The oldest and first written language we have discovered from ancient civilizations was tablets that held accounting records for a bread shop. It has always been this way.
@thsottiaux@rileybrown Please bring Codex to mobile Tibo. Claude let's me use Code in the cloud, but I can only use Work with ChatGPT mobile. It works well, but it isn't Codex.
@OpenAIDevs Built Shipcheck with Astra: a local-first evidence workspace for agent runs. Find unsupported โdoneโ claims, inspect source lines, and export a handoff. Astra handled the build, adversarial review, and tests.
https://t.co/pSgrqKW7uF
@thsottiaux Web only?? No mobile access? Why? It's bad enough we dont get Codex on mobile either and all tasks are local. What am I supposed to do when I lose my laptop and only have mobile devices?
@thsottiaux I can't even access Codex from my phone or tablet either. I'm in the Samsung ecosystem, but please this would be a massive improvement to the workflow.
@thsottiaux please bring Codex to phones/samsung, as well as making chats/tasks in codex save to the cloud automatically. Lost my laptop, along with all the local Codex chats, and I can't keep working from my Phone while waiting for the new m6 mac mini later this month. Thank you
@sama Genuinely excited for Astra, but today was a clusterfuck. Why not just be upfront about the phased rollout? The entire community looks to OpenAI for transparency and communication on the frontier.
@thsottiaux Expose the same primitives to builders: Project State, events, approvals, agent lifecycle, artifacts, and audit history through a supported runtime API. Otherwise every serious agent product must rebuild the continuity layer outside the platform.
@thsottiaux That requires autonomy contracts, not a global permission toggle: explicit tools, files, domains, actions, recipients, budgets, confidence thresholds, approval gates, expiration, audit trails, rollback, and an emergency stop.
@thsottiaux Persistent projects need persistent teams: planner, researcher, builder, tester, and independent verifier. Each role gets a scope and budget; the user can inspect dependencies, and nothing is finished without evidence against explicit acceptance criteria.
@thsottiaux Proactivity should produce fewer interruptions, not more. Give agents an attention inbox that groups work into decision cards: what changed, why it matters, evidence, recommendation, urgency, reversibility, and the smallest decision required from the user.
@thsottiaux Project State should separate facts from intent. Add a collaboration contract: objectives, priorities, risk posture, definition of done, escalation rules, and how the agent should challenge assumptions. Keep it explicit, versioned, and visible when it shapes an action.
@thsottiaux Memory is probabilistic retrieval: what might matter here? Project State is authoritative: what is true for this project now? The first may be fuzzy. The second must be explicit, scoped, versioned, attributable, and reversible. Treating them as the same guarantees drift.
@thsottiaux The missing leap is not merely longer context or a smarter model. It is durable agency: shared state, a clock, a workspace, and bounded authority. Once those exist, chat becomes one interface into an ongoing collaborationโnot the container for it.
@thsottiaux Ship native Project State. Give every Codex project a user-editable, versioned source of truthโgoal, decisions, architecture, backlog, blockers, current status, and next actionsโthat every thread and subagent automatically reads and updates. Context shouldnโt live inside one chat
@thsottiaux Mobile should be a mission-control surface, not a smaller chat box: inspect the plan, agent tree, dependencies, diffs, tests, artifacts, blockers, cost, and status; then approve, redirect, pause, retry, or stop any part of the run.
@thsottiaux The workspace must survive any conversation, model, device, or execution environment. A task should checkpoint locally, resume in the cloud, move to another machine, and return without losing its state, pending work, evidence, or authority boundaries.
@thsottiaux Autonomy should be earned, not granted globally. Let agents run in shadow mode, record what they would have done, compare with user decisions, then graduate specific permissions as reliability is demonstrated. Trust should expand per capabilityโand remain revocable.