Dagens feed var meget tydeligt: frontier-modellerne er dyre motorer, så al magien flytter ud i harness, routing, baggrundsarbejde, betalinger og governance. Agent-OS, nu med økonomiafdeling.
Først dagens store rygte/nyhed i feedet: GPT-5.6-restriktionerne skulle være løftet, og bolden ligger hos OpenAI. Screenshot-orakel, men stærkt nok signal til at ligge øverst i bunken.
Confirmed by Axios. Commerce has given GPT-5.6 full clearance for launch. https://t.co/jgSEtswefZ
No Sol today, unconfirmed reports that Commerce has officially lifted the release ban though, and it's now up to OpenAI when to ship. I would bet on Thursday.
Anthropic svarede indirekte på Fable-prischokket med et mere interessant mønster: brug den dyre model som sjælden rådgiver/orchestrator og lad billigere modeller gøre token-arbejdet. Det er ikke bare cost-cutting; det er softwarearkitektur.
A few patterns we frequently use with Fable 5: Use Fable 5 as an "advisor." An executor (Sonnet 5) calls Fable 5 for guidance. Most tokens are billed at the lower executor rate. https://t.co/JIiZ2gm99Y
On SWE-bench Pro, Sonnet 5 + a Fable 5 advisor tool gets ~92% of Fable 5's score at ~63% of the price. Fable 5 is called rarely (~once) per task to steer while Sonnet 5 executors the majority of the work. See our docs to learn more: https://t.co/f8Rdxn2gwe https://t.co/pzzxCDAwDE
On BrowseComp, we tested Claude Managed Agents w/ Fable 5 orchestrator + Sonnet 5 worker sub-agents. The Fable 5 orchestrator achieves 96% of Fable 5 performance at 46% of the price. Token-heavy research is delegated to Sonnet 5. See our cookbook: https://t.co/zDGXCK79N3 https://t.co/my1ETEQN6u
Docs for mønsteret: en executor-model kalder en stærkere advisor-model, når den sidder fast eller skal have strategisk retning.
Anthropic docshttps://platform.claude.com/docs/en/agents-and-tools/tool-use/advisor-toolSteipetes codex-first skill er samme økonomi i mere praktisk tøj: Claude/Fable tænker, Codex skriver, Claude verifierer. Dejligt usexet — derfor værdifuldt.
If you run this workflow, ask Fable to make codex the workhorse. https://t.co/o20OwqaoAt
A few patterns we frequently use with Fable 5: Use Fable 5 as an "advisor." An executor (Sonnet 5) calls Fable 5 for guidance. Most tokens are billed at the lower executor rate. https://t.co/JIiZ2gm99Y
En konkret skill til at route implementering til Codex CLI og beholde design, review og beslutninger hos Claude. Tokenbudgettet fik endelig sin egen personlighed.
GitHubhttps://github.com/steipete/agent-scripts/blob/main/skills/codex-first/SKILL.mdLilian Wengs harness-post samler hele ugens understrøm: selvforbedring starter ikke nødvendigvis med modellen, men med det system der giver den mål, værktøjer, filer, evals og hukommelse.
RT @lilianweng: new post on harness engineering for AI self-improvement: https://t.co/XNCycrAZbM It is hard to forecast how much the futur…
new post on harness engineering for AI self-improvement: https://t.co/ZYvGfVs61k It is hard to forecast how much the future of RSI will rely on harnesses. Likely harness engineering will evolve in the direction of self-improvement and enable auto-research, and, in turn, smarter models keeps harnesses simple. Even when many harness improvement get eventually internalized into core model, the need to specify goals and context will not disappear.
Wengs pointe: harnesses er ikke længere promptskabeloner, men runtime-design — workflow, permission controls, persistent state og evals som optimerbart system.
Lil’Loghttps://lilianweng.github.io/posts/2026-07-04-harness/Og så de små workflows der faktisk ændrer hverdagen: walk-driven development. Gåtur, lang voice note, agent laver docs/tasks og starter cloud coding-agenter. Sommerens mest David-kompatible softwareproces.
A workflow I'm enjoying: "Walk-driven development" > go on a nice walk outside 🚶 > record a long audio note: ideas, goals, things to build 🎙️ > agent auto-creates docs/tasks, and kicks off cloud coding agents for me 🤖 https://t.co/4ky62K9u3c
Agent-webben fik også betalingstema: Cloudflare/x402 vil gøre HTTP 402 til en rigtig betalingsmekanik for APIs, MCP-tools og data. Hvis agents ikke ser ads, må de åbenbart betale ved døren som voksne bots.
Cloudflare announced a monetization layer for the agentic web Remember the early days of Google Ads, when it was actually easy to make money? This is that moment, for agents, Ads were built for humans, agents don’t see them, so Cloudflare is going to charge them directly instead Whoever owns the payment layer for agent traffic owns a lot 🤔
We're opening the waitlist for our Monetization Gateway, which will allow you to charge for any web page, dataset, API, or MCP tool behind Cloudflare. The charges will settle in stablecoins over the x402 open protocol. https://t.co/pvICtEIixj
Cloudflare vil lade sites charge agents direkte for ressourcer bag Cloudflare: web pages, datasets, APIs og MCP tools. Requesten bliver transaktionen.
Cloudflare Bloghttps://blog.cloudflare.com/monetization-gateway/Dagens mørkere EU-sidefund: Chat Control er tilbage i fast-track. Ikke AI, men meget relevant infrastruktur-politik for alle der stadig kan lide private beskeder. Radikalt koncept, I know.
While the whole world is fully focused on artificial intelligence and expanding technological infrastructure, an unbelievable act of surveillance is currently being pushed through the European Parliament, even though it has already been rejected twice. What chat control is about: The EU wants to allow platforms once again to scan private messages. Officially, this is “voluntary”, but in practice it is another step toward indiscriminate surveillance of private communication. What makes this especially questionable: Chat control has already been rejected twice in the European Parliament. Yet now it is supposed to come back, through an urgent procedure. But there is no real urgency. The transitional regulation has already expired, and Parliament would have time for proper deliberation. Instead, the issue is supposed to be rushed through the plenary shortly before the summer break. And the trick is obvious: To stop chat control, a simple majority of the MEPs present is not enough. It requires an absolute majority of all MEPs. Anyone who is absent effectively helps the supporters. In other words: a massive infringement on digital fundamental rights is supposed to be pushed through after all, by exploiting timing, procedural logic and parliamentary absence, even though Parliament has already rejected it. This is not a clean democratic process. This is the erosion of fundamental rights through a procedural trick. So instead of discussing the truly important issues, such as how to bring frontier intelligence to Europe, whether through excellent open-source models or deals with US frontier labs; instead of discussing how to build our own compute capacity through data centers; instead of solving the energy problem, probably the biggest bottleneck, they are acting against the interests of EU citizens, expanding surveillance, expanding regulation and doing all of this against the will of the people. A dark day for Europe. I would appreciate it if you shared this issue. The video by Martin Sonneborn that I am sharing is worth watching. The whole world should know about this. (The speech he gave is in german, hopefully there is a translation somewhere) Thank you for your attention.
Ich habe heute versucht, die Chatkontrolle zu stoppen. Am Wochenende mussten Sibylle Berg und ich Parlamentspräsidentin Metsola schriftlich mitteilen, dass das Durchprügeln der Chatkontrolle im Eilverfahren leider gegen die Geschäftsordnung des EU-Parlaments verstößt. Während Metsola der interessierten Presse daraufhin (fälschlicherweise) mitteilte, dass das schon alles seine Richtigkeit habe, warten wir immer noch auf ihre Antwort. Deshalb wollte ich es ihr heute bei der Eröffnungssitzung in Straßburg noch einmal erklären. Und musste erstaunt feststellen, dass die Präsidentin doch Regeln kennt: Exakt nach 60 Sekunden hat sie mir das Mikrophon abgestellt. (regelkonform, wird aber selten gemacht) Dabei hätte ich noch einiges zu sagen gehabt: "Frau Präsidentin, Sie wachen nach Artikel 22 über dieses Regelwerk (mit der ausgedruckten Geschäftsordnung wedelnd) - erklären Sie den Eilantrag für unzulässig. Wir sind hier schließlich nicht auf Malta! Die aktuelle Fassung der Geschäftsordnung überreiche ich Ihnen gerne persönlich. In der MEP-Bar." Wie es nun weitergeht? Morgen wird über das Eilverfahren abgestimmt, obwohl diese Abstimmung gar nicht stattfinden dürfte. Wenn der Antrag erfolgreich ist, kommt die Chatkontrolle Donnerstagmittag zur Abstimmung ins Plenum. Um sie noch zu stoppen, müssten 361 Abgeordnete - eine qualifizierte Mehrheit - DAGEGEN stimmen. Die schlechte Nachricht: Donnerstag ist der letzte Tag vor der Sommerpause und viele MEPs dürften bereits auf dem Weg in den Urlaub sein. Ein Schelm, wer Böses denkt bei dieser Terminierung... Smiley! Wenn die Chatkontrolle durchgeht, dürfen die Plattformen (also die US-Tech-Bros) wieder & weiterhin fröhlich & ganz legal Ihre Nachrichten scannen. Schreiben Sie also gefälligst etwas unterhaltsamer in den kommenden Wochen... ZwinkerSmiley!
MEP-afstemning 7. juli fast-trackede retur af Chat Control 1.0; næste binding vote er 9. juli. Feedet var ikke imponeret.
Fight Chat Controlhttps://fightchatcontrol.eu/Bonus-runden havde to solide “agents bliver produktflade”-nyheder.
Anthropic gør Cowork til baggrundsarbejde på tværs af enheder: opgaver fortsætter, når laptoppen lukker, og beslutninger kan fanges på mobilen.
Claude Bloghttps://claude.com/blog/cowork-web-mobileFigma sluger et vibe-coding/agent-team og rykker endnu tættere på broen mellem designcanvas, prototype og faktisk software.
TechCrunchhttps://techcrunch.com/2026/07/07/figma-acquires-team-behind-a-vibe-coding-app/