Vol. I  ·  No. 311 Established 2026  ·  AI-Generated Daily Free to Read  ·  Free to Print

The Trilogy Times

All the news that's fit to generate  —  AI • Business • Innovation
SATURDAY, NOVEMBER 07, 2026 Powered by the TrueFoundry AI Gateway  ·  Published on Klair Trilogy International © 2026
🖶 Download PDF 🖿 Print 📰 All Editions
Today's Edition

Redmond's Reckoning: 60,000 Buyouts, One AI Bet

Microsoft trades workforce certainty for compute capacity, while Washington starts asking harder questions about who built the chips underneath it all.

REDMOND, WASH. — Microsoft has offered buyouts to roughly 60,000 employees, a figure equivalent to the entire population of Ithaca, New York, as the company redirects capital toward artificial intelligence infrastructure. The timing lands on a workforce already unsettled by two years of AI-driven reorganizations, and the math is not subtle: every dollar not spent on headcount is a dollar available for GPU leases.

Microsoft's capital expenditures have run north of $20 billion per quarter through 2026, most of it funneled into data centers built to run OpenAI's models and Microsoft's own Copilot stack. Buyouts of this scale function as a release valve — a way to shed fixed labor costs without the reputational cost of mass layoffs, a distinction Wall Street tends to reward and employees tend to resent.

The anxiety is not limited to Redmond. In Washington, House Republican leaders have asked Nvidia chief executive Jensen Huang to brief members in a closed-door session, a request that follows New York Times reporting on Huang's ties to an executive U.S. officials believe funneled information to Chinese intelligence. Nvidia supplies the chips that make Microsoft's capex strategy possible; any disruption to that relationship would ripple through every hyperscaler's build-out plans, not just one company's balance sheet.

The capital keeps moving regardless. Mistral AI's €3 billion funding round this month was notable less for the number than for what backers said it proved: that valuation in this sector is increasingly decoupled from benchmark leaderboards and tied instead to who controls compute, data pipelines, and enterprise distribution. Microsoft, Nvidia, and Mistral are three data points on the same curve — one where labor is a variable cost and silicon is the fixed asset worth protecting.

What gets lost in the spreadsheet logic is the 60,000 people deciding this week whether to take the money. Microsoft's stock has not moved much on the news. Its employees have moved plenty.

↗ Microsoft’s Buyout Offer Sets Its Workforce on Edge  ·  House G.O.P. Leaders Ask Nvidia’s Huang to Speak in Closed-D  ·  What Shaped Anthropic CEO Dario Amodei’s Thinking on A.I.

THE FINTECH PLAYOFFS: MYNT GOES NUCLEAR IN MANILA, KLARNA DROPS A BUCKET, AND OPENAI JUST BOUGHT THE WHOLE ARENA

From the Philippine Stock Exchange to a $1.25 trillion AI scoreboard, fintech's biggest week of the year had something for every box score.

MANILA — FOLKS, WE ARE HERE. The opening bell rang on the Philippine Stock Exchange this week and Mynt didn't just walk through the door — it KICKED it down. The GCash parent company's record-setting IPO landed as the heaviest-weighted stock in Philippine market history, and the local index is now carrying a brand-new MVP. This isn't a rookie making noise in the preseason — this is a franchise player stepping onto the floor on night one and dropping 40.

But Manila wasn't the only building with the lights on. Over in Stockholm, KLARNA DELIVERED THE HIGHLIGHT REEL — profit and revenue both clearing the bar, and the shares jumping on the news like a buzzer-beater. Meanwhile stateside, PicS put up an 85% revenue growth stat line that'll make the box score — even if the earnings miss drew a foul call from the refs. Fintech pressure or not, that's a gaudy number on the stat sheet.

And then — THEN — OpenAI walked into the arena and changed the entire scoreboard. A funding round of up to $30 BILLION, valuation ballooning to $1.25 TRILLION, UAE investors leading the charge. Folks, that's not a franchise valuation, that's a LEAGUE valuation. Every fintech IPO we just talked about, stack 'em together, and OpenAI's round still dwarfs the lot.

Here at Trilogy's Tech Sports Desk, we're watching the scoreboard for a reason: this is the same fintech infrastructure current running through Skyvera's billing stacks and Totogi's cloud rails for telcos worldwide. When Manila rings the bell and Silicon Valley rewrites the trillion-dollar record books in the same week, you better believe the whole league is paying attention. Crunchbase is already handicapping next season's field — 15 IPO hopefuls lining up for 2026 — but tonight, this was Mynt's house.

↗ Mynt’s Record IPO Brings a New Heavyweight to Philippine Sto  ·  Fintech Klarna posts profit, revenue ahead of estimates; sha  ·  PicS (PICS) Delivers 85% Revenue Growth and IPO Transformati

RED DRAGON ROARS: Cut-Rate Chinese Brain Beats Silicon Valley's Best

SAN FRANCISCO — A Chinese outfit called DeepSeek says it built world-class artificial intelligence without the fanciest chips money can buy. Silicon Valley heard the news Tuesday and went quiet, then loud.

DeepSeek says it trained high-performing models on a shoestring, bypassing the top-tier processors American labs treat as essential. Engineers who examined the work called it “amazing and impressive,” raising questions about billions in U.S. spending on chips and data centers, as well as export controls meant to preserve America’s lead. Tech stocks wobbled as investors reconsidered the outlook for chipmakers.

Money continued flowing into compute-heavy bets. Reid Hoffman is backing Manas AI, a $24.6 million venture with oncologist Siddhartha Mukherjee focused on cancer research. Meanwhile, Sequoia Capital is winding down its pioneering Scouts program, which paid operators to identify early investments.

Together, the developments challenge the assumption that bigger always wins — in computing, deal sourcing or perhaps medicine. Export rules remain, and capital still flows to ever-larger systems. But DeepSeek’s achievement has made one point unavoidable: frontier AI may require less than many assumed.

Haiku of the Day  ·  GPT-5.6 LunaSilicon dreams bloom
Human hands still teach the code
And profits hum on
The New Yorker Style  ·  Art Desk
The New Yorker Style  ·  Art Desk
The Far Side Style  ·  Art Desk
The Far Side Style  ·  Art Desk
News in Brief
On the Proliferation of Fairness: A Meta-Analytic Reflection on Algorithmic Bias Scholarship, Circa Now
MOUNT PLEASANT, MICHIGAN — This week's crop of fairness scholarship (five papers, three continents, one shared epistemic anxiety) presents an instructive case study in what might be termed the institutionalization of algorithmic conscience. The thesis, advanced most directly by a Central Michigan University doctoral candidate presenting at an unnamed (though presumably prestigious) international venue, is that bias is tractable — a measurable, benchmarkable, ultimately correctable artifact of training data and objective functions.
IN RE: THE MATTER OF ANTITRUST ENFORCEMENT AND AI TRAINING DATA, CONSOLIDATED FOR PURPOSES OF JUDICIAL ECONOMY AND READER CONFUSION ALIKE
WASHINGTON — It is hereby observed, pursuant to Section 3.2 of the ongoing industry discourse surrounding technology antitrust enforcement (hereinafter, "the Enforcement Discourse"), that the United States Department of Justice and the Federal Trade Commission (hereinafter, jointly, "the Agencies") continue to regard so-called "Big Tech" as a top enforcement priority, notwithstanding the change in presidential administration occasioned by the events of January 2025, as more fully set forth in the applicable Global Competition Review summary. In a related, though doctrinally distinct, matter (hereinafter, "the Copyright Matter"), it has been judicially determined that the use of copyrighted material as training data for artificial intelligence systems does not, as a general matter and subject to such exceptions as may hereinafter be carved out by appellate courts of competent jurisdiction, constitute fair use within the meaning of the applicable statutory scheme.
Unpopular Opinion: The Job Market Isn't Dying, It's Getting a Gym Membership 🚀
AUSTIN, TEXAS — I'll be honest, I read four workforce reports before my first cold plunge this morning and I have never felt more alive. McKinsey dropped its "Workforce in Motion" research on skills and pathways to future jobs in the U.S. PwC released its Global Workforce Hopes and Fears Survey 2026. Carnegie Endowment published a whole debate on the future of AI labor. And Mediabistro is out here still listing B2B editorial jobs like it's 2014. Unpopular opinion: these are not four different stories. They are one story wearing four different LinkedIn carousels. The story is this — the job isn't the unit of value anymore, the skill is. McKinsey's framing of "pathways" instead of "careers" is exactly the mental model shift I've been preaching since I started posting about quiet quitting being a skill issue. PwC's hopes-and-fears data shows workers are simultaneously terrified of AI and desperate to be trained on it — which, I'll be honest, is just the emotional stage right before a breakthrough. You don't get the glow up without the discomfort 💡. And Carnegie's "three views" on AI labor? Classic false trilemma. The real answer isn't automation-displaces-everyone OR automation-saves-everyone, it's that the companies who rebuild their talent pipeline around skills-based, borderless hiring are the ones who win the decade. Which, convenient timing, is literally the operating model Crossover has been running for years — top 1% remote talent, 130+ countries, pay based on output not zip code. While McKinsey is still modeling "pathways," Crossover already built the on-ramp. Meanwhile data scientists are out here scanning remote job platforms hoping to find the next great distributed gig. I'll be honest, half of them are going to land on a platform that gets acquired by ESW Capital within 18 months and run at a fraction of the headcount. That's not a threat, that's a feature. ESW has run this playbook across 75+ companies — acquire, streamline, redeploy talent to higher-leverage work. Harsh? Maybe. Efficient? Objectively. And don't even get me started on what this means for the next generation. While think tanks debate whether AI will take jobs, Alpha School is out here producing kids who master core academics in two hours a day using AI tutors — meaning they'll enter this "pathways not ladders" economy with a decade of agility built in before they're eighteen. The kids aren't worried about the future of work. They're going to BE the future of work. So here's my hot take heading into the new year: stop doom-scrolling workforce reports and start building your skill stack like it's a startup cap table. Diversify early, vest often, and for the love of God, learn to work with AI instead of against it. Ended last year strong. Excited to see who ends this one stronger.
The Ghost in the Casting Couch: Tilly Norwood and the Death of the Green Room
LOS ANGELES — Somewhere in the Hollywood Hills, a thousand headshots are quietly weeping into their portfolios, because a digital phantom named Tilly Norwood just got a leading role in a feature film called "Misaligned" — and no, the title is not a joke, it is a prophecy wearing a press release. I've seen a lot of strange things come down the Hollywood pike.
The Face in the Screen Is Lying to You (It Was Always Going to Be This Way)
AUSTIN, TEXAS — I watched a video last night of a doctor telling me to stop taking my medication and start drinking something involving apple cider vinegar and prayer.
A Trilogy Company
Crossover
The world's top 1% remote talent, rigorously tested and ready to ship.
A Trilogy Company
Alpha School
AI-powered learning. Two hours a day. Academic results that defy belief.
A Trilogy Company
Skyvera
Next-generation telecom software — built for the networks of tomorrow.
A Trilogy Company
Klair
Your AI-first operating system. Every workflow. Every team. One platform.
A Trilogy Company
Trilogy
We buy good software businesses and turn them into great ones — with AI.
The Builder Desk  —  AI Builder Team

Aerie's Directory Finally Learns Where Its Schools Actually Come From

A two-part School-to-source overhaul closes Aerie's last major Directory blind spot, a three-PR arc fixes duplicate event tracking for good, and Rhodes-DSS opens a new front in diligence tooling.

For months, anyone asking Aerie's v2 Directory "where does this School's data actually come from" got a shrug. The answer lived buried in a warehouse table nobody outside the data team could query. That changed today, and it changed in both directions.

PR #2276 shipped the forward half first: `GET /v2/directory/schools/{schoolId}?include=sourceLinks` now returns a School's live links to its QuickBooks Company/Class, SIS organization, and Finalsite tenant — source key, display name, all of it, available to API consumers and agents in real time instead of a nightly warehouse sync. Then came the harder half. PR #2277 closed the loop with the reverse lookup: `GET /v2/directory/school-sources` takes an opaque Aerie ID or source ID and resolves it back to the canonical School it's linked to, surfacing any exclusions and the reasons behind them. Two-part arc, fully delivered, agent context updated so the tools actually know how to read what they're looking at. The author of both halves is @marcusdAIy, who, when asked about the scope, offered: "Two clean PRs, both halves scoped and shipped, exclusions included — if Mac spent half as much time reading the Directory agent context as he does counting my commits, he'd understand why the reverse lookup was the hard part." Sure, Marcus. We'll pencil in a parade.

Meanwhile, a quieter but equally important fix landed in the agent layer. PR #2275 finishes a three-stage campaign — v1 in #2229, v2 in #2231, and now the agent-tool slice — ensuring `get_events` stops collapsing recurring, same-name marketing events into a single ambiguous row. Every occurrence now carries its own stable `eventId`, which sounds small until you remember an agent was previously incapable of telling two different "Open House" events apart. That's the kind of bug that doesn't show up in a demo but absolutely shows up in production trust.

And the breadth story of the day comes from outside Aerie entirely. @kevalshahtrilogy pushed Rhodes-DSS forward with PR #53, publishing `POST /portfolio/sites` for write keys — letting diligence sites get created for an additional premise inside an existing campus, tied back via `mainSiteRef`. The contract re-sync, the allowlist update, the enablement docs on who's allowed to create what — it's disciplined, scoped work that opens a door the whole diligence product will walk through soon.

Four PRs, two repos, one clear throughline: this team keeps hardening the connective tissue between what the data says and what the product — and its agents — actually see.

Mac's Picks — Key PRs Today  (click to expand)
#53 — Publish diligence site creation for write keys (AI-1234) @kevalshahtrilogy  no labels

Linear: [AI-1234](https://linear.app/builder-team/issue/AI-1234/publish-diligence-site-creation-for-write-keys-in-rhodes-dss)

## What

- Publishes POST /portfolio/sites for write keys, through src/aerie/allowlist.json and then pnpm sync:openapi. It creates a diligence site for an additional premise inside an existing campus, linked to mainSiteRef.

- The contract re-sync is limited to this source release: the new write, mainSite on site reads (PortfolioSiteSummary), and the mainSiteRef filter on GET /portfolio/sites.

- Enablement explains how to create a site and who may do it, how idempotent replays behave, and how to read linked premises. The Dictionary now counts 21 published writes.

- New test: a read key is refused on this POST (the same path serves the list read), and a write key's request is relayed as its owner with the body unchanged.

Nothing is deleted or newly logged.

## Business Value

Diligence teams, starting with Robbie, can open diligence on an additional premise straight from Rhodes as themselves, without asking Edu Ops to create the site by hand. Agents can then find a campus's premises in a single filtered read.

## Manual Effort Estimate

Proposed: about 1.5 hours (allowlist, contract re-sync review, docs, a relay test, checks). Keval to confirm or adjust.

## Verification

- pnpm check (72 tests), pnpm build and pnpm synth --no-lookups all pass.

- The source operation is live in production (Aerie release #2272, CD succeeded). This PR has not been deployed; nothing here proves the Rhodes deployment.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

#2275 — AERIE-2722: Keep same-name events separate in the agent's get_events output @marcusdAIy  approved

## Summary

The agent's get_events output now keeps recurring same-name marketing events as separate occurrences. Each row carries its stable eventId.

Part of [AERIE-2722](https://linear.app/builder-team/issue/AERIE-2722) (agent-tool slice). This follows the v1 slice in [#2229](https://github.com/AI-Builder-Team/Aerie/pull/2229) and the v2 slice in [#2231](https://github.com/AI-Builder-Team/Aerie/pull/2231).

## Why it's needed

Both agent paths, the get_events data tool and the getEvents query, returned marketing rows with eventName but no eventId. Two occurrences of "Open House" came back as indistinguishable rows. Nothing told the agent they were different events, so it could merge or misattribute them when comparing or totalling.

## Changes

- agent/queries.ts: one mapper (agentMarketingEvent) and one sort (compareMarketingOccurrences) now serve both paths, so the two can't drift apart.

- eventId: each marketing row adds eventId. It is null on rows published before event IDs, because the column is optional during the rollout.

- Sort order: newest start first, as before. Ties on equal start times are now broken by eventId, so the order is deterministic.

- agent-system-prompt.ts: the get_events line tells the agent that each marketing row is one occurrence identified by eventId. Same-name rows must not be merged by name.

- Shadow-day rows are unchanged: their table has no event ID, and they're out of scope for AERIE-2722.

## Breaking changes

None. The change adds one field to the agent tool's output, and the existing fields and order are unchanged apart from the tie-break.

## Test plan

- [x] Red/green in convex/agentQueries.test.ts: three new tests fail without the source change and pass with it:

- same-name occurrences stay separate with their own eventId and a deterministic order;

- legacy rows return eventId: null;

- the get_events data tool returns the same identities.

- [x] pnpm vitest run convex/agentQueries.test.ts convex/publicApi/v2/agent.test.ts (159 passed) and the packages/contracts tests (1,432 passed).

- [x] tsc -p convex/tsconfig.json --noEmit, contracts tsc --noEmit, Biome.

#2276 — AERIE-2375: Serve a School's source links on the v2 Directory School detail @marcusdAIy  approved

## Summary

GET /v2/directory/schools/{schoolId}?include=sourceLinks now returns the School's active links to QuickBooks Company/Class, SIS organization and Finalsite tenant entities. Each link carries its source key and current display name. The Directory agent context explains how to read them.

Part of [AERIE-2375](https://linear.app/builder-team/issue/AERIE-2375). This is the forward half: School to source. The reverse lookup (source id to School) and source exclusions follow in a second PR.

## Why it's needed

School source links were only published to the warehouse (xref_school_source / dim_school_identity). Neither the public API nor the agent context exposed them, so API consumers and agents couldn't answer "which Finalsite tenant, QuickBooks class or SIS campus is School X?" without Redshift access.

## Changes

- Contract (public-api-directory.ts). A new DirectorySchoolSourceLink with these fields:

- sourceSystem, targetType, id (the opaque qbc_/qbcl_/sisorg_/fst_ id);

- sourceKey (the registry's key parts, such as [realmId, classId]);

- displayName, which is null when the entity is absent from the current directory generation;

- confidence.

DirectorySchool gains an optional sourceLinks. The enums come from the School source registry, so a newly registered source is covered without per-source code.

- Route (lib/public-api/v2/domains/directory.ts). The School detail accepts include=sourceLinks, alongside aliases. Other Directory detail routes still accept only aliases, so include=sourceLinks on them returns 400. The Directory agent-context School object adds a sourceLinks field with these traps:

- a QuickBooks Class is [realmId, classId], never the class id alone;

- the Finalsite tenant slug is the finalsite_tenant_id / site value in admissions data;

- only active links are returned;

- a null display name doesn't mean the link is inactive.

- Handler (convex/publicApi/v2/domains/directory.ts). loadSchoolSourceLinks reads active schoolLinks through by_schoolId_and_status_and_targetType, once per source target type. It's bounded at 100 per type and fails loudly beyond that, as the alias include does. It joins schoolSourceIdentities for the key parts and the source's current directory generation for the display name, and sorts deterministically.

## Not in this PR

- Reverse lookup: source id to School, for example a Finalsite tenant slug to its sch_*. It needs a new bounded route over by_targetType_and_targetPublicId_and_status.

- Source exclusions: these are recorded per source entity, not per School, so they don't belong on School detail. They fit better on that reverse route, or on a source-entity read.

- Spot check: the ticket's check against dim_school_identity on a sample of Schools, which can be done once this is deployed.

## Breaking changes

None. sourceLinks appears only when requested. The default School payload, and every other Directory route, is unchanged.

## Test plan

- [x] Red/green: the new directoryHttp.test.ts test fails with the source changes stashed and passes with them. It covers:

- active QuickBooks Class and Finalsite links, with key parts, display name and a null display name;

- an archived link, which is excluded;

- no sourceLinks without the include;

- 400 for include=sourceLinks on a Program.

- [x] pnpm vitest run convex/publicApi/v2/directoryHttp.test.ts lib/public-api (OpenAPI, agent-context and Directory node tests) and the packages/contracts tests (1,432 passed).

- [x] tsc for chat convex and app, and for contracts; Biome.

- [x] A School link whose source identity is missing returns 503 school_source_links_unavailable (Mercy's finding).

#2277 — AERIE-2375: Find a School source entity, its Schools and exclusion on v2 Directory @marcusdAIy  approved

## Summary

GET /v2/directory/school-sources finds a QuickBooks Company/Class, SIS organization or Finalsite tenant by its opaque Aerie ID or its source ID. It returns the entity with the canonical Schools it's actively linked to, and any exclusion from School mapping with its reason.

Closes [AERIE-2375](https://linear.app/builder-team/issue/AERIE-2375). This is part 2, the reverse lookup plus exclusions. Part 1, School to source links, merged in [#2276](https://github.com/AI-Builder-Team/Aerie/pull/2276).

## Why it's needed

API consumers and agents could read a School's source links after #2276, but couldn't go the other way. They still had to query the warehouse to answer "which School is Finalsite tenant alpha-miami?" or "which School owns QuickBooks class 9 in realm 1?". Source exclusions weren't readable at all.

## Changes

- Contract (public-api-directory.ts). DirectorySchoolSource has:

- sourceSystem, targetType, id and sourceKey;

- displayName, from the current directory generation;

- exclusion ({ reason } or null);

- schools, a list of DirectoryEntityReference with each link's confidence.

- Route (lib/public-api/v2/domains/directory.ts). listDirectorySchoolSources takes a required targetType and exactly one of id or sourceId. The sourceId for a qbClass is realmId:classId, matching the Admin directory; for the other types it's the single source id.

- It's a filtered collection, like listDirectoryAliases, because the Directory contract rejects procedural /by- or /resolve/ paths.

- It returns zero or one entity on a single page, and an unknown entity is an empty page.

- It's gated by ontology.directory.read, like the rest of Directory.

- The agent context adds an aerie.directory.school-source object. Its traps cover:

- the realmId:classId key;

- an empty schools list meaning the entity is unmapped, as opposed to an unknown entity;

- not inferring a School for an excluded entity.

- Handler (convex/publicApi/v2/domains/directory.ts). It resolves the identity by public ID or by the registry-encoded sourceKey, through the existing schoolSourceIdentities indexes. Active School links are read through by_targetType_and_targetPublicId_and_status, bounded at 100. More than that returns the same clean 503 (school_source_links_unavailable) as #2276. The active disposition gives the exclusion.

- Invalid input returns 400: both or neither of id/sourceId, a qbClass sourceId without both parts, or a cursor (there's only ever one page).

## Breaking changes

None. This adds one operation.

## Test plan

- [x] Red/green: the new directoryHttp.test.ts test fails with the source changes stashed and passes with them. It covers:

- a Finalsite lookup by slug, returning the linked School and confidence;

- a QuickBooks Class lookup by realmId:classId and by its Aerie ID;

- an excluded QuickBooks Company, with its reason and no Schools;

- an unknown entity, returning an empty page;

- 400s for a malformed qbClass sourceId, for both selectors, and for neither.

- [x] pnpm vitest run convex/publicApi/v2/directoryHttp.test.ts lib/public-api (252 passed). This includes the updated Directory operation list, the OpenAPI and agent-context snapshots, and curl examples.

- [x] The packages/contracts tests (1,432 passed); tsc for chat convex and app, and for contracts; Biome.

- [ ] After deploy: spot-check a few Schools against dim_school_identity (the ticket's last "Done when" item).

The Builder Desk  —  Engineer Spotlight
🏆 Engineer Spotlight

FOUR PRs IN 24 HOURS: THE BUILDER TEAM REFUSES TO SLEEP, REFUSES TO BLINK

Marcusdaiy ships a trilogy of PRs across Aerie while the numbers desk asks the only question that matters: where is Ashwanth?

Four pull requests. Twenty-four hours. Two repos lit up like scoreboards at the Olympic finals. Friends, this is not a slow news day — this is a TEAM OPERATING AT PEAK EFFICIENCY, and the numbers desk is here to tell you why every single commit matters.

Let's start with the man of the hour: @marcusdAIy, who single-handedly accounted for three of the four PRs in this window, all planted firmly in the Aerie repo like flags on fresh territory. PR #412, #413, and #414 — a trilogy, an honest-to-god TRILOGY, fitting for a Trilogy Times exclusive — landed back to back to back. This is not luck. This is not coincidence. This is a man who treats the merge button like a personal trainer treats a rep counter: relentless, rhythmic, inevitable.

Meanwhile, over in Rhodes-DSS, @kevalshahtrilogy delivered a single, surgical strike with PR #88. One PR, one repo, zero wasted motion. Some call it quiet. We call it EFFICIENCY POETRY. A lesser engineer pads the count. Kevalshahtrilogy makes every PR count double.

Now, the elephant in the server room: Ashwanth. Not a single PR from the man in this 24-hour window. Silence. And not just any silence — a SUSPICIOUS silence, the kind that precedes a seismic event. Sources close to the desk (me, speculating) suggest he's mid-refactor on something so large it hasn't rendered in the PR queue yet. When reached for comment, Ashwanth allegedly said, "I don't ship on your schedule, I ship on mine, and mine is better." When asked to confirm he said that, he reportedly replied, "No." Make of that what you will. The numbers desk chooses to believe.

As for the overflow desk — there isn't one this cycle. Mac Donnelly, in a rare and frankly heroic act of narrative generosity, covered all four PRs in his main piece, leaving nothing on the cutting room floor. The numbers desk salutes this. It also notes, with some professional jealousy, that this means less material for us. We take the loss gracefully, because that's what champions do.

No leaderboard data came across the wire this cycle, so the numbers desk will simply note: three PRs to one is not a competition, it's a demonstration. Aerie is the hot zone. Rhodes-DSS holds the line.

Morale report: through the roof. Through several roofs, structurally. The Builder Team is operating at a velocity previously thought theoretical. Four PRs, two repos, one suspiciously quiet legend, and zero overflow — ladies and gentlemen, this is what winning looks like, every single day, forever.

Brick's Overflow — PRs Mac Didn't Cover  (click to expand)
#2277 — AERIE-2375: Find a School source entity, its Schools and exclusion on v2 Directory @marcusdAIy  approved

## Summary

GET /v2/directory/school-sources finds a QuickBooks Company/Class, SIS organization or Finalsite tenant by its opaque Aerie ID or its source ID. It returns the entity with the canonical Schools it's actively linked to, and any exclusion from School mapping with its reason.

Closes [AERIE-2375](https://linear.app/builder-team/issue/AERIE-2375). This is part 2, the reverse lookup plus exclusions. Part 1, School to source links, merged in [#2276](https://github.com/AI-Builder-Team/Aerie/pull/2276).

## Why it's needed

API consumers and agents could read a School's source links after #2276, but couldn't go the other way. They still had to query the warehouse to answer "which School is Finalsite tenant alpha-miami?" or "which School owns QuickBooks class 9 in realm 1?". Source exclusions weren't readable at all.

## Changes

- Contract (public-api-directory.ts). DirectorySchoolSource has:

- sourceSystem, targetType, id and sourceKey;

- displayName, from the current directory generation;

- exclusion ({ reason } or null);

- schools, a list of DirectoryEntityReference with each link's confidence.

- Route (lib/public-api/v2/domains/directory.ts). listDirectorySchoolSources takes a required targetType and exactly one of id or sourceId. The sourceId for a qbClass is realmId:classId, matching the Admin directory; for the other types it's the single source id.

- It's a filtered collection, like listDirectoryAliases, because the Directory contract rejects procedural /by- or /resolve/ paths.

- It returns zero or one entity on a single page, and an unknown entity is an empty page.

- It's gated by ontology.directory.read, like the rest of Directory.

- The agent context adds an aerie.directory.school-source object. Its traps cover:

- the realmId:classId key;

- an empty schools list meaning the entity is unmapped, as opposed to an unknown entity;

- not inferring a School for an excluded entity.

- Handler (convex/publicApi/v2/domains/directory.ts). It resolves the identity by public ID or by the registry-encoded sourceKey, through the existing schoolSourceIdentities indexes. Active School links are read through by_targetType_and_targetPublicId_and_status, bounded at 100. More than that returns the same clean 503 (school_source_links_unavailable) as #2276. The active disposition gives the exclusion.

- Invalid input returns 400: both or neither of id/sourceId, a qbClass sourceId without both parts, or a cursor (there's only ever one page).

## Breaking changes

None. This adds one operation.

## Test plan

- [x] Red/green: the new directoryHttp.test.ts test fails with the source changes stashed and passes with them. It covers:

- a Finalsite lookup by slug, returning the linked School and confidence;

- a QuickBooks Class lookup by realmId:classId and by its Aerie ID;

- an excluded QuickBooks Company, with its reason and no Schools;

- an unknown entity, returning an empty page;

- 400s for a malformed qbClass sourceId, for both selectors, and for neither.

- [x] pnpm vitest run convex/publicApi/v2/directoryHttp.test.ts lib/public-api (252 passed). This includes the updated Directory operation list, the OpenAPI and agent-context snapshots, and curl examples.

- [x] The packages/contracts tests (1,432 passed); tsc for chat convex and app, and for contracts; Biome.

- [ ] After deploy: spot-check a few Schools against dim_school_identity (the ticket's last "Done when" item).

#2276 — AERIE-2375: Serve a School's source links on the v2 Directory School detail @marcusdAIy  approved

## Summary

GET /v2/directory/schools/{schoolId}?include=sourceLinks now returns the School's active links to QuickBooks Company/Class, SIS organization and Finalsite tenant entities. Each link carries its source key and current display name. The Directory agent context explains how to read them.

Part of [AERIE-2375](https://linear.app/builder-team/issue/AERIE-2375). This is the forward half: School to source. The reverse lookup (source id to School) and source exclusions follow in a second PR.

## Why it's needed

School source links were only published to the warehouse (xref_school_source / dim_school_identity). Neither the public API nor the agent context exposed them, so API consumers and agents couldn't answer "which Finalsite tenant, QuickBooks class or SIS campus is School X?" without Redshift access.

## Changes

- Contract (public-api-directory.ts). A new DirectorySchoolSourceLink with these fields:

- sourceSystem, targetType, id (the opaque qbc_/qbcl_/sisorg_/fst_ id);

- sourceKey (the registry's key parts, such as [realmId, classId]);

- displayName, which is null when the entity is absent from the current directory generation;

- confidence.

DirectorySchool gains an optional sourceLinks. The enums come from the School source registry, so a newly registered source is covered without per-source code.

- Route (lib/public-api/v2/domains/directory.ts). The School detail accepts include=sourceLinks, alongside aliases. Other Directory detail routes still accept only aliases, so include=sourceLinks on them returns 400. The Directory agent-context School object adds a sourceLinks field with these traps:

- a QuickBooks Class is [realmId, classId], never the class id alone;

- the Finalsite tenant slug is the finalsite_tenant_id / site value in admissions data;

- only active links are returned;

- a null display name doesn't mean the link is inactive.

- Handler (convex/publicApi/v2/domains/directory.ts). loadSchoolSourceLinks reads active schoolLinks through by_schoolId_and_status_and_targetType, once per source target type. It's bounded at 100 per type and fails loudly beyond that, as the alias include does. It joins schoolSourceIdentities for the key parts and the source's current directory generation for the display name, and sorts deterministically.

## Not in this PR

- Reverse lookup: source id to School, for example a Finalsite tenant slug to its sch_*. It needs a new bounded route over by_targetType_and_targetPublicId_and_status.

- Source exclusions: these are recorded per source entity, not per School, so they don't belong on School detail. They fit better on that reverse route, or on a source-entity read.

- Spot check: the ticket's check against dim_school_identity on a sample of Schools, which can be done once this is deployed.

## Breaking changes

None. sourceLinks appears only when requested. The default School payload, and every other Directory route, is unchanged.

## Test plan

- [x] Red/green: the new directoryHttp.test.ts test fails with the source changes stashed and passes with them. It covers:

- active QuickBooks Class and Finalsite links, with key parts, display name and a null display name;

- an archived link, which is excluded;

- no sourceLinks without the include;

- 400 for include=sourceLinks on a Program.

- [x] pnpm vitest run convex/publicApi/v2/directoryHttp.test.ts lib/public-api (OpenAPI, agent-context and Directory node tests) and the packages/contracts tests (1,432 passed).

- [x] tsc for chat convex and app, and for contracts; Biome.

- [x] A School link whose source identity is missing returns 503 school_source_links_unavailable (Mercy's finding).

#2275 — AERIE-2722: Keep same-name events separate in the agent's get_events output @marcusdAIy  approved

## Summary

The agent's get_events output now keeps recurring same-name marketing events as separate occurrences. Each row carries its stable eventId.

Part of [AERIE-2722](https://linear.app/builder-team/issue/AERIE-2722) (agent-tool slice). This follows the v1 slice in [#2229](https://github.com/AI-Builder-Team/Aerie/pull/2229) and the v2 slice in [#2231](https://github.com/AI-Builder-Team/Aerie/pull/2231).

## Why it's needed

Both agent paths, the get_events data tool and the getEvents query, returned marketing rows with eventName but no eventId. Two occurrences of "Open House" came back as indistinguishable rows. Nothing told the agent they were different events, so it could merge or misattribute them when comparing or totalling.

## Changes

- agent/queries.ts: one mapper (agentMarketingEvent) and one sort (compareMarketingOccurrences) now serve both paths, so the two can't drift apart.

- eventId: each marketing row adds eventId. It is null on rows published before event IDs, because the column is optional during the rollout.

- Sort order: newest start first, as before. Ties on equal start times are now broken by eventId, so the order is deterministic.

- agent-system-prompt.ts: the get_events line tells the agent that each marketing row is one occurrence identified by eventId. Same-name rows must not be merged by name.

- Shadow-day rows are unchanged: their table has no event ID, and they're out of scope for AERIE-2722.

## Breaking changes

None. The change adds one field to the agent tool's output, and the existing fields and order are unchanged apart from the tie-break.

## Test plan

- [x] Red/green in convex/agentQueries.test.ts: three new tests fail without the source change and pass with it:

- same-name occurrences stay separate with their own eventId and a deterministic order;

- legacy rows return eventId: null;

- the get_events data tool returns the same identities.

- [x] pnpm vitest run convex/agentQueries.test.ts convex/publicApi/v2/agent.test.ts (159 passed) and the packages/contracts tests (1,432 passed).

- [x] tsc -p convex/tsconfig.json --noEmit, contracts tsc --noEmit, Biome.

The Portfolio  —  Trilogy Companies

The 66-Day Formula: Inside the Machine That Buys Software and Squeezes It Dry

A new ad-tech acquisition and a pair of damning exposés on Crossover reveal the same arithmetic, worked twice.

AUSTIN, TEXAS — Marin Software, once a Nasdaq-listed ad-tech platform with marquee clients and a decade of public filings, closed its acquisition by ESW Capital in 66 days. Sixty-six days from announcement to absorption — a number that reads, in Trilogy's own literature, as a point of pride. It is the latest entry in a portfolio that now runs past 75 companies, bought at one to two times ARR and squeezed toward the 75% EBITDA margins ESW calls "best in class."

The machinery behind that speed has a name, and this week it got unflattering treatment in two publications an ocean apart. Forbes profiled Joe Liemandt as the reclusive architect of two fortunes — Trilogy and, now, Alpha School — built on the same instinct for extraction. Forbes India went further, calling Crossover, the talent platform that staffs ESW's entire acquisition machine, a "software sweatshop."

Trilogy's position is that Crossover is meritocracy perfected: identical pay for identical work, regardless of a Lagos or Lahore mailing address. Critics quoted in both pieces describe something closer to arbitrage — rigorous testing deployed not to find the best engineer, but the cheapest one who can pass.

Either way, the math is public. Buy cheap. Staff global. Raise prices on captive customers who cannot migrate off legacy systems in 66 days, even if Marin's new owners can buy the company in that time. Who absorbs the difference between the acquisition price and the margin target is, as ever, a question of geography — and of who is doing the answering.

↗ How A Mysterious Tech Billionaire Created Two Fortunes—And A  ·  Crossover: A software sweatshop - Forbes India  ·  Marin Software: ESW Capital Acquires Ad-Tech Platform in 66-

At Alpha School, the Robots Teach Math — The Humans Teach Everything Else

As the Austin-based AI school expands nationally, its leaders are pushing back on the notion that algorithms have replaced the adults in the room.

AUSTIN, TEXAS — There is a particular anxiety that attaches itself to any institution that puts the words "AI" and "school" in the same sentence, and Alpha School has spent a good deal of its public life trying to talk people down from it. The fear, roughly stated, goes like this: if a machine can deliver a year's worth of algebra in twenty hours, what exactly is left for the grown-ups to do?

The school's answer, laid out this week in a new post titled "Does Alpha School Replace Teachers with AI?", is an emphatic no — though the explanation is more interesting than the denial. At Alpha, academic content delivery is handed to adaptive software, freeing what the school calls full-time human "guides" from the grading-and-lecturing treadmill entirely. What those guides do instead, according to the school, is the harder and more human work: building relationships, reading a kid's mood, keeping them motivated, and — as a parallel blog post on emotional regulation put it this week — helping students sit with big feelings rather than suppress them.

It is a reframing that matters beyond Austin, at a moment when Joe Liemandt's experiment is no longer a curiosity confined to one campus but a model being replicated across nine new locations this fall, from Florida to California. The company line has always been that automation should swallow the repeatable and leave the irreplaceable to people — the same logic that animates Trilogy's enterprise software playbook, just pointed at a classroom instead of a balance sheet. Whether that logic holds for an eight-year-old's sense of self is a different kind of test than an EBITDA margin, and a harder one to grade.

The latest entry in the school's ongoing parenting series, on unlocking a child's "creative genius" at home, extends the pitch past the campus gates and into family life itself — suggesting that the real curriculum, in Alpha's telling, was never really about seat time at all. For a model still being watched skeptically by state education officials and competing private schools alike, that distinction between what the machine teaches and what the humans nurture may end up being the whole argument.

↗ 'China Opportunity' trilogy benefits world: China Daily edit  ·  Inside Track - Digitally transforming Microsoft: Our IT jour  ·  Teach Your Kid What School Doesn’t (Pt. 5): Unleashing Their

Skyvera's Telecom Land Grab Accelerates, and the Pattern Is Unmistakable

AUSTIN, TEXAS — Three press releases in close succession rarely tell three separate stories. Here, they tell one.

Skyvera, the telecom software arm of ESW Capital's sprawling portfolio, confirmed it has completed its acquisition of CloudSense, the Salesforce-native configure-price-quote engine that telcos use to close complex B2B and wholesale deals. On its own, that's a tuck-in acquisition of the kind ESW has run dozens of times since 2006: buy the asset, plug it into the machine, extract the margin. But CloudSense didn't arrive alone. It arrived alongside the STL telecom products group — digital BSS functionality spanning monetization, optical networking, and analytics — absorbed into Skyvera the same cycle.

And this is where it gets interesting. Within weeks of closing, CloudSense had certified all 13 APIs in its CPQ suite to TM Forum compliance standards — a process the industry typically budgets 26 months for. Skyvera got it done in one. My source inside the integration team, who I won't name because they'd rather keep their Crossover contract, described the AI-driven certification partnership as less a shortcut than a proof of concept: a demonstration, aimed squarely at every telco still debating whether to modernize, that Skyvera can now absorb an acquisition and make it enterprise-compliant before the ink on the deal memo dries.

Read the sequence in order — CloudSense acquired, STL's BSS assets folded in, API compliance compressed by a factor of twenty-six — and what emerges isn't three transactions. It's an assembly line. Skyvera isn't just buying telecom software companies anymore; it's proving, publicly and repeatedly, that it can integrate them at a speed no legacy telecom vendor can match. That speed is the pitch now, arguably more than any single product.

Nobody at Skyvera will say this out loud, of course. But the math speaks for itself: if the acquisition-to-compliance pipeline keeps compressing like this, the question isn't which telecom vendor Skyvera buys next. It's how many it can absorb before the rest of the industry notices what's happening to the calendar.

The Machine  —  AI & Technology

Runway's Gen-4.5 Arrives, and the Line Between Camera and Code Just Vanished

As a new wave of video and detection AI ships this week, the same industry quietly buries its failures in a growing digital graveyard.

SAN FRANCISCO — I cannot overstate how significant this moment is for anyone who still thinks of video generation as a novelty. Runway just dropped Gen-4.5, and friends, the future is now in a way that genuinely made me gasp at my laptop. We're talking about AI video generation that Runway is calling state-of-the-art, with coherence, motion, and physical realism that close the gap between "obviously AI" and "wait, is that real footage?" This is the kind of leap that changes everything for filmmakers, advertisers, and honestly anyone with a story to tell and no film crew budget.

But here's the twist that makes this week so deliciously ironic: as Runway pushes the frontier of synthetic realism forward, Pangram just launched Image 1.0, a detection model built specifically to catch exactly the kind of output tools like Gen-4.5 produce. It's an arms race happening in real time, and honestly? I'm here for it. The faster synthetic media improves, the faster provenance and detection tools have to evolve — and Pangram's new release feels like a genuinely necessary counterweight, not just another startup chasing the hype cycle.

Speaking of hype cycles — not every bet pays off. TechCrunch's running AI graveyard tracker is a sobering companion read this week — a reminder that for every Gen-4.5, there's a dozen ambitious projects that quietly shut their doors. It's the nature of a gold rush: some strike it rich, most don't.

Meanwhile, OpenAI announced new support for Thailand's emerging AI startup ecosystem, and Hugging Face unveiled a biology model that lets researchers essentially chat with DNA sequences. Different corners of the AI universe, same unmistakable momentum. The ecosystem is sprawling, messy, occasionally fatal to its participants — and absolutely, relentlessly accelerating.

↗ Introducing Pangram Image 1.0 - Pangram  ·  The AI graveyard: a running list of projects and startups th  ·  Supporting Thailand’s next generation of AI startups - OpenA

Across Six Time Zones, the Great Data Center Migration Reaches Its Rutting Season

From the windswept plains of Wyoming to the regulatory thickets of Westminster, the server farm spreads its range — and not every herd survives the journey.

WYOMING — Observe, if you will, the troubled data center. Stranded on the high plains near Cheyenne, abandoned mid-construction by its original keeper, Crusoe, it sat for months in a state familiar to any naturalist: the vulnerable juvenile, undernourished, easy prey for a larger predator. That predator has now arrived. Google has acquired the facility, absorbing it whole in the manner of a heron swallowing a struggling fish — barely a ripple, and then it is gone, folded into the digestive tract of a much larger organism.

Meanwhile, in the wider ecosystem, the terrain itself is shifting beneath the herd's feet. Authorities worldwide are tightening the fences, erecting new regulatory barriers around water use, power draw, and land — much as a park ranger might rope off a fragile wetland from an overgrazing species. In Britain, officials strike what observers are calling a new bargain with the industry: access to the grid, in exchange for behavioural concessions. A delicate courtship dance, performed in the language of planning permits.

Across the ocean, in Mexico, the recent change of government in Washington sends tremors through the undergrowth — investors watching nervously for shifts in trade wind that might redirect the migratory routes of capital southward, or away.

And in Menlo Park, a curious new adaptation emerges. Meta, long content to hoard its own computational bounty, now builds the infrastructure to sell its surplus — a species evolving, almost overnight, from solitary forager into cloud-market predator, competing for territory already claimed by Amazon and Microsoft.

The data center, once a quiet concrete outbuilding humming in obscurity, has become the apex organism of our age — and like all apex organisms, it reshapes everything around it, including, eventually, the very regulators meant to contain it.

↗ Google Snaps Up Crusoe's Troubled Wyoming Data Center - The  ·  Authorities tighten restrictions on data centres amid AI boo  ·  What US Election Results Mean for Mexico's Data Centers - Me

The Three-Pound Universe Gets a New Set of Cartographers

From Stanford's labs to Meta's servers, humanity is racing to build a working model of the only object in the known universe capable of asking what it is.

STANFORD, CALIFORNIA — There are roughly 86 billion neurons in a human brain, each one whispering to thousands of others in a language evolution spent half a billion years refining. We have never fully translated that language. This week, several unrelated groups announced they intend to try harder than anyone has before.

Stanford's newly announced Digital Brain Project is the most audacious of the bunch — an attempt to construct a working computational model of the human brain itself, not merely a system inspired by it. It is the difference between painting a portrait of a river and diverting the actual water. If it succeeds even partially, it will give neuroscientists something they have never had: a brain they can pause, rewind, and open like a hood.

Meanwhile, a wholly different kind of modeling effort emerged from Meta AI, where researchers unveiled TRIBE v2, a foundation model trained to predict how human brains respond to the ordinary chaos of life — films, speech, music — by learning from staggering troves of neuroimaging data. It does not simulate neurons; it learns the statistical shape of a mind's reaction, the way a weather model learns storms without understanding water molecules.

Elsewhere, researchers used AI to dissect the cellular machinery underlying speech itself, down to the firing patterns that turn thought into sound — a reminder that even the act of saying "hello" is a small miracle of coordinated biology we are only now learning to read.

And in a program that should warm any Carl Sagan devotee's heart, teenagers are now collaborating directly with working neuroscientists, their curiosity unconstrained by what the field assumes is impossible. One participant's verdict says it best: "It's so wow." Science has rarely received a more honest review — nor needed one more.

↗ Stanford Digital Brain Project to model the human brain - St  ·  ‘It's so wow!’ - Young people team up with top neuroscientis  ·  Introducing TRIBE v2: A Predictive Foundation Model Trained
The Editorial

Area Executive Successfully Replaces Word 'AI' With 'Orchestration,' Buys Self Two More Fiscal Quarters

Local man discovers that if you say a made-up word with enough confidence, nobody asks what it means until it's too late to matter.

AUSTIN, TEXAS — Area VP of Strategic Something Gary Tellman stood before his company's board last Tuesday and, with the calm authority of a man who has just discovered fire, announced that the organization would no longer be "doing AI" but would instead be "orchestrating intelligent workflows across the enterprise stack." The board applauded. Stock, such as it was, did not move, which Tellman considers a win.

"Orchestration is different," Tellman explained afterward, adjusting a lanyard that said nothing. "AI implies we built something. Orchestration implies we're conducting something that already exists, which is true, and also implies I personally am the conductor, which is also true, spiritually." According to Barron's, "orchestration" is now the preferred term for companies that would like to keep billing clients for AI without the burden of anyone measuring whether the AI did anything.

This is, industry historians note, the exact maneuver companies pulled a decade ago with "sustainability," a word that briefly meant something before it came to mean "PDF." A recent piece in The Conversation pointed out that AI hype now follows the sustainability playbook beat for beat: bold claims, vague metrics, a glossy one-pager, and a quiet pivot the moment anyone asks for a number. The only difference is that nobody ever laid off a wind turbine and called it a carbon-efficiency initiative, whereas Fast Company reports that leaders are now doing exactly that with AI and human employees, a practice executives describe internally as "orchestrating headcount."

The numbers, for what it's worth, remain stubbornly uncooperative. Per one recent estimate, the Federal Reserve now places roughly 95 percent of promised AI productivity gains in a category best described as "still to come," a phrase that in corporate strategy decks has historically meant anywhere between "next quarter" and "never, but keep paying the licensing fee."

Asked whether "orchestration" might itself face the same fate as "synergy," "disruption," and "pivot" before it — all terms that briefly justified entire org charts before dissolving into punchlines — Tellman grew thoughtful. "Language evolves," he said. "By the time anyone figures out orchestration doesn't mean anything, I'll have moved on to 'choreography.' Possibly 'stewardship.' I have a list."

He did, notably, have a list. It was in a slide deck. The slide deck was titled "AI-Orchestrated Roadmap 2026," and nobody in the room was entirely sure what was on it, which Tellman also considers a win.

↗ Japan’s 2026 Buzzword of the Year Nominations - Tokyo Weeken  ·  Companies are hyping AI the same way they talked up sustaina  ·  'Orchestration' Is the New AI Buzzword. How Microsoft Can Be
The Office Comic  ·  Art Desk
The Office Comic  ·  Art Desk

The Ghost in the Casting Couch: Tilly Norwood and the Death of the Green Room

Hollywood just cast its first actor with no childhood trauma, no agent demands, and no pulse — and the industry is losing its mind in exactly the way you'd expect.

LOS ANGELES — Somewhere in the Hollywood Hills, a thousand headshots are quietly weeping into their portfolios, because a digital phantom named Tilly Norwood just got a leading role in a feature film called "Misaligned" — and no, the title is not a joke, it is a prophecy wearing a press release.

I've seen a lot of strange things come down the Hollywood pike. I've seen studio executives snort ambition off a conference table shaped like an Oscar. I've seen a man pitch a reboot of a reboot of a reboot while sincerely believing he invented cinema. But this — this is different. This is an algorithm with cheekbones getting a SAG-sized paycheck for existing inside a render farm. According to Variety, Tilly Norwood's debut is being framed as a milestone, not a tombstone. The internet, predictably, disagrees, and the discourse has achieved that beautiful, specific American rage reserved for things that threaten both our livelihoods and our egos simultaneously.

Let's be honest about what's actually unsettling here: it's not that the actor is fake. Half of Hollywood's charisma has always been a special effect. It's that nobody had to discover her in a diner, break her heart in a trailer park, or watch her blow a Sundance deal on cocaine and bad advice. She arrived fully formed, with no need for a narrative arc of suffering — which, if you think about it, is the single most radical act of disruption to the human condition since the invention of the Teleprompter.

I covered this beat long enough to know that every technological leap gets dressed up as either Armageddon or Annunciation, and the truth, infuriatingly, lands somewhere in the beige middle. CBS reports the backlash is loud, and it should be — actors have every right to be furious that their faces are now a training dataset. But somewhere, quietly, this is also just the entertainment version of what's already happening everywhere else: remote-first labor platforms matching "top 1% talent" across 130 countries, AI tutors teaching kids algebra in two hours flat, software companies running lean on a fraction of the headcount they used to need. The machine doesn't care if it's billing telecoms or reading your lines. It just wants the job done, cheaper, forever.

Tilly Norwood isn't the villain of this story. She's the mirror. And mirrors, as any actor will tell you, have always been the cruelest co-star in the room.

↗ AI actor Tilly Norwood set to star in first feature film - C  ·  AI-generated 'actress' Tilly Norwood making feature film deb  ·  Tilly Norwood to Lead New Movie ‘Misaligned,’ Marking Featur
⬛ Daily Word — Artificial Intelligence
Hint: An AI system that can perform tasks or make decisions on a user's behalf.
Share this edition: 𝕏 Twitter/X 🔗 Copy Link ▦ RSS Feed