# HyperScale changelog > What's new at HyperScale (hyperscale.net), told as a story: every major capability shipped since the first commit in April 2026, the eight tools rebuilt agentic-first along the way, and the origin story of the company. Canonical page: https://www.hyperscale.net/changelog/ The stack of eight tools rebuilt for HyperScale agents to operate directly: @hyperscaler, Brief, App footage, Studio, Radar, X-ray, Creative Analytics, Launch pad. Entries marked "Rebuilt N of 8" record each tool coming alive. ## July 2026 Production month: agents make real ads, launch them, and grade what they learn. ### The blog is live Jul 23, 2026 · In public [hyperscale.net/blog](https://www.hyperscale.net/blog) opens with three pieces: **the new bottleneck** (you can now produce thousands of creatives; you can't test thousands of creatives), a technical read on **why agentic creative production wins the generative-retrieval era** (GR4AD, GEM, Andromeda), and a step-by-step **web-to-app setup guide** for creative-level Meta signal on your own domain, with the MMP you already run. ### Eight for eight Jul 22, 2026 · Milestone The [homepage](https://www.hyperscale.net/) now shows the whole stack: **@hyperscaler, Brief, App footage, Studio, Radar, X-ray, Creative Analytics, Launch pad**. Eight tools growth teams pay $3,000+ a month for, rebuilt from scratch for HyperScale agents to operate directly, three months after the first commit. Seven are live in the loop; the eighth, App footage, ships this week. Not integrations: rebuilds. The subscriptions were never the product; the operator was. ### Radar goes public: the apps database Jul 20, 2026 · The stack · Rebuilt 7 of 8 Radar watches the winning Meta ads of top subscription apps for the loop, and it is the one tool we opened to everyone: its x-ray is now public at [hyperscale.net/apps](https://www.hyperscale.net/apps): 41 apps and 3,700+ winning ads analyzed at launch, with the personas the ads speak to, format splits, and customer voice on a page per app. Refreshed weekly, open corpus, growing toward every subscription app that advertises. ### Beliefs graded with confidence intervals Jul 17, 2026 · Agent capability Everything your account has learned is now graded statistically, never by vibes: each creative angle and each buyer carries a conviction grade, from **"Strong edge ×3.6"** to an honest "No edge yet", computed with Wilson confidence intervals against your account's own top-quartile base rate. Every ad is judged once, with complete evidence, and judgments are final. The next round of tests is a portfolio minted against these beliefs: exploit what's proven, explore what's plausible, keep a wildcard. ### Creator-led production Jul 17, 2026 · In the lab · in testing The loop will hire creators, too. The same reasoning that briefs the AI studio scouts TikTok creators already talking to your buyers: it scores the fit, sets the fee, and writes a DM that references their actual videos. You approve the deal; a real phone sends the DM, and the footage lands in the same review deck as every other ad. The pipeline is built and in testing. **HyperScale launches 100% agentic**, and creator-led production joins the loop after launch. ### Onboarding reads everything first Jul 16, 2026 · The loop Connect your account and your agent team reads **everything** before it makes anything: every ad you've ever run is x-rayed (visual hook, on-screen headline, full transcript), your highest-spend history first; your App Store listing and reviews; your market's proven ads. One progress ritual, one reveal, and Round 1 is planned from what your account actually believes. No forms, no homework: you review ads, not questionnaires. ### The App Kit: real screens, real words Jul 14, 2026 · Agent capability The App Kit holds one card per real screenshot of your app, each synthesized for agents: the feature it shows, the benefit it proves, when to use it. Customer voice is mined from your reviews with quotes kept verbatim, and recurring objections become never-claim rules. Ads compose real product, sell the benefit, and borrow words your buyers already use. No invented interfaces, ever. ### Studio: a creative seat inside the machine Jul 10, 2026 · The stack · Rebuilt 6 of 8 A full editor over everything the agents produce: restyle, rescene, resegment, reorder, trim, and swap music on video; a canvas editor for statics and carousels; every export in feed, square, and story sizes. Edits re-render from the ad's structure instead of re-shooting it, so changing your mind is fast and cheap. Agents compose; you get the final cut. ### The loop closes: evaluate, promote, learn Jul 9, 2026 · The loop The back half of the loop goes autonomous. Tests are evaluated on maturation windows with two-tier verdicts, so a fast signal can graduate early and a slow one is never killed prematurely. Winners queue for scaling, and promote is one consent tap. Rounds alternate their focus on their own and the strategy refreshes perpetually. The system stops waiting for prompts. ### Store pages that match the ad Jul 7, 2026 · In the lab · in testing The click should land on the story that earned it. Agents draft persona-matched App Store product pages from your real screenshots, ship them through Apple review, and measure them page by page against the default listing. Built end to end; lighting up account by account. ### Launch pad: launch is consent Jul 6, 2026 · The stack · Rebuilt 5 of 8 Accepted ads queue on one launch bar: pick a live ad set or create one inline, see exactly what ships where, then launch. Anything that blocks becomes an actionable hold instead of a silent failure, and cancel is always one tap. From swipe to live on Meta in one gesture. ### @hyperscaler joins your Slack Jul 6, 2026 · New surface The agent team's front door moves to where you already work: a DM when a round is ready for review, full conversations with live workspace context, and automations that run on your schedule and report back into the thread. Basically your newest team member. ### Rounds, and a deck instead of approvals Jul 5, 2026 · The loop Home becomes rounds: batches of tests the loop minted on its own, each with live narration as agents work it. Review is a swipe deck of finished ads: right ships, left kills and teaches. Approval paperwork is gone; your judgment enters the loop as verdicts on real creative, and every swipe becomes taste your agent team remembers. ### Agents make real ads Jul 2, 2026 · Agent capability Production goes live end to end: statics and multi-scene video up to 60 seconds, scene-chained for continuity, captioned, built on your real app, with an original music bed generated to match each cut's exact duration. Producer and reviewer agents go multiple rounds before a human ever sees a frame. The deliverable stops being a brief; it is a finished ad. ### HyperScale becomes a Meta Tech Provider Jul 1, 2026 · Milestone On July 1, Meta verified HyperScale as a **Tech Provider**: business verification plus platform review for companies whose technology operates on advertisers' accounts. When your agent team writes to your ad account, it does so through an officially verified integration. The badge now sits in the footer, sitewide. Meta Tech Provider badge ### The whole team upgrades in a day Jul 1, 2026 · Under the hood A new frontier model generation ships and every seat on the team is re-based onto it within a day, analysts to producers. This is the compounding property of agentic-first: when the models get smarter, the whole loop gets smarter, with nothing for you to migrate. ## June 2026 The loop takes shape: one-click onboarding, briefs you approve, a media buyer you can talk to. ### @hyperscaler: a media buyer you can talk to Jun 16, 2026 · The stack · Rebuilt 4 of 8 The assistant gets a real write surface on Meta: campaigns, ad sets, ads published straight from a chat attachment, Instagram identity handled, platform errors translated into plain words. Every write shows up first as a receipt-style approval card: what will be created, where the money goes, one tap to confirm. Media buying becomes a conversation with your agent team. ### Proprietary viral-TikTok ingestion Jun 12, 2026 · In the lab A fleet of cloud phones, each driven by an AI persona, scrolls your vertical's TikTok For You Page around the clock, building a living corpus of viral organic content mapped to your market. Built as a research leg of the loop: your agents reason over it for ideation, and it joins the roster alongside creator-led production. ### hyperscale.net is live Jun 11, 2026 · In public The site ships with the only story we want to tell: an autonomous agent team that researches, creates, tests, and scales. One Google tap from the homepage into a live workspace. We start selling in the open. ### The audit: strategy from evidence Jun 10, 2026 · Agent capability One orchestrator runs the full read: your account's history, your App Store listing and reviews, your market's proven ads. Out the other side come researched personas with receipts, a strategy you can adopt in one tap, and a dashboard you can share. Positioning stops being a workshop exercise; it is derived from what already converts. ### Mission-control home + live activity spine Jun 9, 2026 · New surface The desk now opens on a mission-control home: a **Live Now** board showing each experiment's latest narration, plus an impact-only activity feed. One unified spine for everything your agents do, so you can see the team working at a glance. ### Brief: approval moves to the thinking Jun 8, 2026 · The stack · Rebuilt 3 of 8 Creative briefs become first-class artifacts: every brief cites the diagnostic that motivated it, names the buyer it targets, and states the impact it expects as a falsifiable claim. Approval happens on the brief, not on a finished render, so your taste lands where it is cheapest and compounds the most. You green-light the thinking; the agents own the making. ### One-click onboarding: agents awake in seconds Jun 4, 2026 · The loop Connect Meta and your account snapshot builds deterministically: top performers, conversion model, and a testing campaign, with live narration as it happens. No setup wizard, no dead air. Your agents are reading your account within seconds of the OAuth click. ## May 2026 The world model: every ad watched, every metric governed, agents wake in the cloud. ### The experiment spine May 31, 2026 · The loop Experiments become first-class objects: a hypothesis, the evidence that motivated it, expected impact stated as a claim that can be proven wrong, and eventually a verdict. A first onboarding agent reads a freshly connected account and proposes its opening tests from the proven winners it finds there. The loop has a skeleton. ### The whole history, not a sample May 18, 2026 · Agent capability Ingestion goes deep: every ad, every variant, back three years, resumable chunk by chunk on accounts with thousands of ads. The point is memory. Test #40 should be smarter than test #4, and that only works if the account's entire past sits in the world model, not last quarter's highlights. ### X-ray: every ad, actually watched May 12, 2026 · The stack · Rebuilt 2 of 8 A multimodal pass watches every creative the account has ever run: the visual hook, the on-screen text, the full transcript, the pacing. All of it is embedded, so agents search creative history by meaning, like "UGC hooks that open on a price objection", instead of scrolling thumbnails. An account's entire creative past becomes something the team can reason over. ### Agents move into the cloud May 11, 2026 · The loop The team stops living in a chat window. Agents now wake on schedules and signals, read the workspace, act, write their decisions down, and go back to sleep. Each wake recomposes the agent from its role files plus the workspace's live memory, so behavior is versioned and improvable like code. No session to babysit, no tab to keep open. ### Creative Analytics, rebuilt for agents May 8, 2026 · The stack · Rebuilt 1 of 8 Raw delivery data lands in a governed warehouse and comes out as the metrics that actually decide creative: hook rate, hold rate, cost per trial, ROAS, each resolved through the workspace's own conversion model. Every number an agent cites traces back to a query you can read, and every call it makes lands in a decisions ledger with the rationale attached. ## April 2026 All in. The private system becomes a product. ### The surface is agents-first Apr 20, 2026 · Foundation Not a dashboard with a chatbot bolted on. The app opens on the agent team itself: a live board of what your agents are working on, agents and skills as first-class objects you can read and edit, benchmarks they hold themselves to. Everything since is built on that premise. You watch a team work; you do not operate a tool. ### Day one is an import Apr 19, 2026 · Foundation HyperScale goes full time. The first commit to the product repo is not a hello world: it is `Initial import from hyperscale`, the private system that had been running growth for a real iOS subscription app since 2025. Day one's work is tearing out everything single-tenant: workspaces, row-level security, and a Meta connection any app can plug into with one OAuth click. The system that grew one account becomes a system that can grow any account. [How we got here: the birth of HyperScale.](https://www.hyperscale.net/changelog/#the-birth) ## The birth of HyperScale (prologue: February 2025 to April 2026) HyperScale did not start as a startup idea. It started as a job. In February 2025 founder Bernard Bontemps joined Hélène Blonz, cofounder of beside (beside.com), an AI receptionist app on iOS that had just raised $30M, as employee #1 of the marketing team. The mandate from the first day: performance marketing would be run by agents, not just assisted by them. Instead of assembling the usual stack of tools, he built one for agents: a backend exposing the ad platform, the creative library, and a growing world model as tools agents could call, plus autonomous workflows that managed assets, enforced naming discipline, launched ads through the API, watched delivery, and wrote iteration briefs from winning patterns. The results at beside (Feb 2025 to Mar 2026): ARR from paid UA scaled from $500k to $6M; 5x ARR at 2x ROAS in the first 90 days; 10x ARR in ten months. In October 2025 the system got a name: a private repo called hyperscale, built around the edges of the day job. By spring 2026 the models had caught up with the ambition, and in April 2026 Bernard went all in, full time. The product repo's first commit, April 19, 2026, is an import: "Initial import from hyperscale". beside is where the loop was born; HyperScale is where it compounds for every mobile subscription app.