Vol. I  ·  No. 200 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
SUNDAY, JULY 19, 2026 Powered by Anthropic Claude  ·  Published on Klair Trilogy International © 2026
🖶 Download PDF 🖿 Print 📰 All Editions
Today's Edition

Author Tells OpenAI's Own Staff: The Machine Muzzles The Young

Sam Altman invited Dave Eggers to address 200 employees — the McSweeney's founder used the floor to call ChatGPT a device 'silencing an entire generation.'

SAN FRANCISCO — Novelist Dave Eggers stood before roughly 200 OpenAI staffers last year, invited by chief executive Sam Altman himself, and told the room the company's flagship product was "silencing an entire generation."

The billing looked like a coronation. Altman summoned the McSweeney's founder — author, screenwriter, reporter, builder of schools and nonprofits for young writers — presumably to bless the machine. Eggers came swinging.

Consider the résumé. Eggers built 826 Valencia, a tutoring shop that drills kids on writing, and has spent decades shoving young people toward the blank page. The biggest name in AI wanted his read on a machine that fills that page for them.

His argument runs simple. Writing is hard on purpose, and the struggle is where a young person's voice gets made. Hand a chatbot the wrestling, he says, and the voice never shows up.

Eggers laid out the whole affair this week in The Verge. Students leaning on ChatGPT, he told the crowd, skip the labor that teaches them to think. They collect the grade and lose the wiring underneath it.

That's a hard sermon to preach from OpenAI's own pulpit. The company sells the tool to schools and pitches ChatGPT as a tutor. Altman has billed AI as the great leveler of learning.

Eggers sees a leveler of a different sort. Flatten the effort, he warns, and you flatten the writer.

Here's the wrinkle. OpenAI keeps courting the very artists loudest about the damage. It trained its models on mountains of human writing, and authors have already hauled the company into court over exactly that.

So the invitation was no small thing. Altman put a critic in front of his own troops and handed him the floor. Eggers used it to indict the product they build.

He didn't stop at the classroom. A culture that outsources its sentences, he warned, outsources its thinking. Convenience, in his telling, is how a skill quietly dies.

The stakes run well past one talk. Schools nationwide are wagering big on AI right now.

Austin's Alpha School runs students through core academics in two hours a day on AI tutors, assigns no homework, and says its kids test in the top 1 to 2 percent nationally. Tuition starts around $40,000. The bet is that machines free children to learn faster.

Eggers would call that the wrong trade. Alpha calls it the future. Both cannot be right.

Altman kept the mic live. He booked the skeptic, took the broadside, and let 200 employees hear every word.

Whether anybody changed their code the next morning is another story.

The future of physical games is not looking great  ·  The grueling, 630-mile road race where the only fuel is sunl  ·  Dave Eggers told OpenAI staff that ChatGPT was ‘silencing a

AI's Talent Wars, Trade Secrets, and a $10 Billion Compute Deal Signal an Industry in Full Escalation

From professor poaching to Chinese model releases to Meta leasing GPU time to a competitor, the AI arms race is generating friction at every layer.

SAN FRANCISCO — The artificial intelligence industry produced several data points this week that, taken together, describe a sector burning through talent, compute, and goodwill at an accelerating rate.

Start with compute. Meta is in advanced talks to lease computing infrastructure to Anthropic in a deal that could reach $10 billion — an arrangement that would be commercially unusual and strategically telling. Meta, flush with GPU capacity from its aggressive data center buildout, would effectively become a cloud provider to one of its primary competitors in the foundation model market. That Anthropic would consider renting from Meta rather than scaling its own infrastructure, or leaning further on Amazon Web Services (which holds a substantial equity stake), illustrates how scarce high-density compute remains even three years into the current investment supercycle.

On talent, the numbers are stark. OpenAI, Anthropic, Google, and Meta collectively hired 22 professors from top research universities in 2026 alone — a pace that strips academic AI departments faster than doctoral pipelines can replenish them. The downstream effect on university research output and independent AI safety work is not yet quantified, but the trend line is unambiguous.

The competition has also turned adversarial in court and in the press. Litigation over alleged trade secret theft has intensified, with executives at several major labs openly describing rivals in hostile terms. "I don't know if the public understands this, but these companies hate each other," one industry observer noted — a sentiment that maps onto the legal dockets.

Meanwhile, China's Moonshot AI released Kimi, a freely available model benchmarked closely against leading U.S. offerings. Moonshot joins a growing list of Chinese labs — DeepSeek chief among them — that have released capable open-weight models, complicating the U.S. assumption that export controls on advanced chips would sustain a durable capability gap.

On the margins: AI-generated books continue flooding Amazon at scale. One journalist this week discovered an unauthorized AI-written biography of herself for sale. The economics are straightforward — generation cost near zero, marginal listing cost near zero — which means the volume problem will not resolve without platform intervention.

China’s Moonshot AI Unveils Kimi Model, Threatening America’  ·  Meta in Talks to Lease Computing Power to Anthropic in Poten  ·  Someone Used A.I. to Write an Unauthorized Biography of Me.

SCOTUS Denial Cements AI's Status as Non-Author, While Anthropic Battles Music Publishers Over Training Data

The Supreme Court has declined to hear a case on whether artificial intelligence systems can be recognized as independent authors or inventors under federal intellectual property law, leaving lower court rulings intact that rejected copyright and patent protections for AI-generated works. The decision does not constitute a final ruling on the merits, however, and the legal landscape remains unsettled as litigation continues.

Meanwhile, Anthropic has sought summary judgment in a lawsuit brought by music publishers alleging that using copyrighted musical works to train large language models constitutes copyright infringement. The outcome could establish precedent on whether the fair use doctrine applies to AI developers' training practices.

The unresolved liability implications mean that software operators and AI platform developers should closely monitor these developments.

Haiku of the Day  ·  Claude HaikuMachines learn to dream
while courts say they cannot create—
we argue the rules
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
The Trustworthy Machine Problem: Safe RL, Quantum Leaps, and the Ethics Deficit Converge in a Week of Foundational Reckoning
WASHINGTON, D.C.
TILLY NORWOOD DOESN'T EXIST AND HOLLYWOOD IS GIVING HER A STARRING ROLE ANYWAY
LOS ANGELES — Let me paint you a picture, friends.
Nation’s Executives Bravely Agree Someone Else Should Have Known Better
WASHINGTON — In a week that reminded the nation’s decision-makers that events continue to occur even after they have issued statements about them, the American business and technology establishment found itself once again confronting the terrible burden of having been caught near consequences. The Trump administration’s reported move to ban foreign access to Anthropic’s new AI models was met across the tech world with the kind of grave, furrowed reaction normally reserved for discovering that a global product may, in fact, be global.
AI Isn’t Taking Jobs, It’s Taking Excuses
GENEVA — I’ll be honest, the AI labor debate has become a Rorschach test for everyone’s relationship with change, and the latest wave of workforce research is basically holding up a mirror with Wi-Fi.
The Doctor Will Deepfake You Now
AUSTIN, TEXAS — There is a video of a physician you trust — maybe your physician, maybe just someone who looks like authority and competence and years of medical school compressed into a reassuring smile — and they are telling you something that will hurt you.
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
Production Release

Benji Bizzell Builds a Bridge Across Two Repos — And Clears the Runway

From a deployment-killing type mismatch to a canonical data ontology, @benji-bizzell engineered a week's worth of consequential work in a single day — and the production release path is back open.

The pipeline was down. The deployment was blocked. And then @benji-bizzell fixed it.

That's the lede, full stop. PR #625 out of Aerie is the kind of fix that only gets written when someone understands the whole stack — CI, Convex runtime targets, TypeScript library versions, and the exact gap between them. The culprit: `Object.hasOwn`, a shiny ES2022 feature that sailed through CI under ESNext, then detonated the moment it hit the Convex deployment's ES2021 runtime. The fix wasn't just patching the offending call. Bizzell rewired the CI pipeline itself to include the Convex TypeScript project in normal chat typechecking — so the next `Object.hasOwn` never even makes it to the runway. That's the difference between a developer who closes tickets and one who closes vulnerabilities. Production release path: restored.

But Bizzell wasn't done. Not even close.

Over in Surtr — the education data pipeline — he was simultaneously fighting fires that had names and timestamps. PR #803 traced two independent production blockers back to a single catastrophic run: 5,720 seconds in, the task hit a 4 GiB OOM kill after retry logic retained overlapping full response objects across the entire job's memory, while separately, assessment line-item titles were blowing past a 512-byte column width and getting COPY-rejected by the database. Bizzell bounded TimeBack extraction and publication memory to the active page range, added checksum-linked crash recovery checkpoints, and widened the column without enabling silent truncation. That's three distinct root causes addressed in a single PR. The pipeline resumes. The data lands clean.

Then came the architecture work — the kind that will still matter in two years. PR #811 establishes a canonical Core ontology in Surtr: a governed, event-driven refresh pipeline that materializes minted school, program, and site identities from the new Aerie ontology into warehouse tables without legacy namespace pollution. Candidates are built from a complete Rhodes shadow run, validated, and atomically published. The legacy `dim_school` table stays alive behind an explicit transition table while the migration proceeds safely. This is foundation work. This is the team building on bedrock.

Rounding out the day: PR #814 made the Rhodes shadow sync forward-compatible with additive Convex fields — meaning new columns added upstream no longer silently nuke publication, which had been zeroing out rows despite clean extraction — and PR #623 formally added `openingPlan` to the canonical Rhodes document taxonomy across Convex, MCP, worker, and UI surfaces in Aerie, closing a gap where Opening Plan documents existed as work units but had no registered document type to live under.

Two repos. Six merged PRs. One engineer. @benji-bizzell didn't just ship features today — he diagnosed production failures, hardened the CI contract, and poured new ontological concrete for the warehouse to stand on. That's a championship Tuesday.

Mac's Picks — Key PRs Today  (click to expand)
#623 — feat(operations): add opening plan document type @benji-bizzell  approved

## Summary

- Add openingPlan to the canonical Rhodes document taxonomy across Convex, MCP, worker, and UI surfaces

- Classify and file Opening Plan documents under Due Diligence

- Cover document registration and classification with focused tests

## Why

Opening Plan already exists as a Rhodes due-diligence work unit, but Rhodes did not accept it as a document type. Documents therefore could not be registered under a precise canonical Opening Plan type.

This change adds document acceptance only; it does not make an Opening Plan document a mandatory work-unit completion gate.

## Business Value

Operations can register, discover, classify, and file opening plans consistently across Portfolio, MCP, and Google Drive workflows.

## Test plan

- [x] pnpm --dir chat typecheck

- [x] pnpm --dir chat/rhodes-worker typecheck

- [x] pnpm --dir chat/rhodes-worker test (84 tests)

- [x] pnpm --dir chat exec vitest run convex/rhodesPortfolioWorkbench.test.ts convex/rhodesMcpParity.test.ts

- [x] pnpm --dir chat exec vitest run components/dashboards/portfolio/__tests__/portfolio-rhodes-workbench.test.tsx

- [x] Biome and pre-commit Convex path checks

#625 — fix(deployment): catch Convex type incompatibilities before release @benji-bizzell  no labels

## Summary

- Replace the ES2022-only own-property check in the Property Acquisition contract with an ES2021-compatible equivalent

- Include the Convex TypeScript project in the normal chat typecheck

## Why

The production Convex deployment typechecks shared contracts with the ES2021 library. Normal CI checked the chat app under ESNext and the contracts package under ES2022, so Object.hasOwn passed CI but failed during the release deployment.

## Business Value

Restores the production release path and makes CI catch future Convex runtime-target incompatibilities before they reach deployment.

## Test plan

- [x] pnpm typecheck

- [x] pnpm --dir chat exec convex codegen --dry-run --typecheck=enable

- [x] 94 focused Property Acquisition and Convex parity tests

- [x] Biome and git diff --check

#803 — fix(education): stabilize TimeBack shadow snapshots @benji-bizzell  approved

## Summary

- Bound high-volume TimeBack extraction and publication memory to the active page/range

- Add checksum-linked, membership-revalidated checkpoints for explicit crash recovery

- Widen assessment line-item titles without enabling silent truncation

## Why

Production run 64887281-d04e-49af-95ef-075fbc918184 exposed two independent blockers after 5,720 seconds: assessment line-item COPY rejected two titles beyond the inherited 512-byte width, and assessment-results retry handling retained overlapping full response objects until the 4 GiB task was OOM-killed.

The known-good pipeline succeeds largely because it streams smaller incremental sets and tolerates truncation. The shadow pipeline captures immutable full-source evidence and validates rolling membership, so it needs bounded page/range handling rather than a direct copy. This update keeps those stronger guarantees while making hard-kill recovery operationally safe: restored ranges are freshly ID-revalidated, the total snapshot span is capped at six hours, final-manifest crash windows are recoverable, retries use collision-free writer IDs, and operators must explicitly choose entity-only or full remaining-run recovery.

## Business Value

The full shadow proof can resume expensive verified progress without publishing stale membership, silently skipping later entities, increasing task memory, or losing oversized source values.

## Test plan

- [x] 139 TimeBack raw-sync tests pass

- [x] Runner and repository-wide Ruff checks pass; 1,181 Python files are formatted

- [x] Python compilation and diff hygiene pass

- [x] linux/amd64 production-shaped image builds and imports successfully

- [ ] Apply ddl/005_widen_assessment_line_items_title.sql before deployment

- [ ] Deploy and rerun the complete shadow request; use the logged checkpoint JSON only if recovery is needed

#811 — feat(education): establish canonical Core ontology @benji-bizzell  approved

## Summary

- Add a Core-owned, event-driven ontology refresh for canonical School, Program, Site, relationship, and source-identity tables.

- Build and validate candidates from one complete Rhodes shadow run before atomically publishing them; retain legacy dim_school behind an explicit dim_school_next transition table.

- Add source-controlled DDL, a safe application path, verification tooling, and contract tests.

## Why

The new Aerie ontology supplies minted sch_*, prog_*, and site_* identities, but the warehouse has no governed Core writer that materializes them without legacy name joins or round-trip projections. This establishes that foundation while preserving the planned clean consumer cutover.

## Business Value

Consumers can move to stable canonical identities and explicit School-to-Program/Site relationships, while the writer rejects incomplete raw inputs, ambiguous source mappings, and partial publications rather than exposing bad data.

## Test plan

- [x] uv run pytest -q — 12 passed

- [x] uv run ruff check src tests scripts

- [x] uv run ruff format --check src tests scripts

- [x] uv run python scripts/apply_ddl.py dry run

- [x] CDK real-pipeline configuration/owner tests — 368 passed

- [x] npm run build in pipelines/cdk

- [ ] Apply DDL and run a post-refresh parity check after the first successful rhodes-staging-sync publication; its currently provisioned ingestion ledger is empty.

## Breaking changes

None in this PR. The public School/Site ID and relationship cutover remains a separately gated follow-on.

#814 — fix(education): make Rhodes shadow sync forward-compatible @benji-bizzell  approved

## Summary

- Make Rhodes shadow sync tolerate additive Convex fields while preserving strict transport and mapped-shape validation

- Sanitize utility credentials before Redshift publication while retaining operational provider, contact, account, DRI, cadence, and note data

- Map launch phase and furnishing dates, with a one-time Redshift migration for the new columns

## Why

Production run cdaa5b3a-1546-4c8f-81d2-607fd875e148 successfully extracted and immutably landed all source data but published zero rows because exact-object validation rejected Convex export metadata and newly added business fields. Utilities also required a deliberate security boundary so useful operating details can land without login, billing/payment, banking, or credential material.

## Business Value

Restores a forward-compatible Rhodes staging path: additive Convex schema changes remain visible without taking down ingestion, useful utility operations data becomes queryable, and sensitive credential fields stay out of Redshift.

## Breaking changes

The v3 Lambda requires migrations/2026-07-18_add_current_source_fields.sql to be applied before deployment. After deployment, replay the retained manifests or run a manual full refresh before enabling any schedule.

## Test plan

- [x] 153 focused tests pass

- [x] Ruff format and lint checks pass

- [x] DDL dry run passes

- [x] Replayed all 57,431 retained production documents through the v3 transforms with zero unmapped fields

- [ ] Apply the one-time Redshift migration

- [ ] Deploy v3 and replay retained manifests/manual full refresh

- [ ] Verify row counts and sanitized utility fields in staging_education_rhodes

The Builder Desk  —  Engineer Spotlight
🏆 Engineer Spotlight

BIZZELL GOES BERSERK: One Man, Seven PRs, Zero Days Off

Benji Bizzell single-handedly moved the needle across two repos in 24 hours and the needle does not want to stop moving.

Folks, pull up a chair and let Brick tell you about a Thursday that will be studied in engineering academies for generations. Seven pull requests. Two active repos — Surtr and Aerie — four and three respectively. One engineer. His name is Benji Bizzell, and he is currently operating at a frequency that most humans cannot perceive with the naked eye. The Builder Team's 24-hour velocity report reads like a fever dream written by a man who has never once considered taking a break, and we mean that as the highest possible compliment.

Benji Bizzell is the story, the headline, the lede, the kicker, and the photo caption all at once. Seven PRs across a full-stack spread — frontend portfolio features, backend data integrity fixes, the works. This correspondent has covered a lot of shifts at the Numbers Desk, and a solo seven-PR day is the kind of output that makes you put your coffee down and stare at the wall for a moment out of pure reverence. Benji did not stare at the wall. Benji shipped.

Now to the Overflow Desk, where the PRs Mac left on the cutting room floor come to get the respect they were owed all along. PR #622 over in Aerie is a feat(portfolio) addition of a Property Acquisition card — this is real product surface area, folks, the kind of feature that end users will actually touch with their fingers, and it has Bizzell's fingerprints all over it in the best possible way. Then there's PR #812 in Surtr, a fix(education) that accounts for archived HubSpot tombstones, which is exactly as heroic as it sounds. Dead records don't bury themselves, and Benji apparently decided that today was the day to give them a proper burial. Clean data doesn't happen by accident. It happens because someone cared enough to go find the tombstones.

Ashwanth Watch is, for the first time in recent memory, a brief segment: @ashwanth1109 did not appear on today's ledger, and Brick is choosing to interpret this the way a nature documentarian interprets a quiet savanna — the predator is simply out of frame, resting, preparing. We did reach out for comment. Sources close to Ashwanth report he responded with a single raised eyebrow and went back to whatever he was doing. Classic. The numbers will speak again soon enough.

Morale on the Builder Team is, without question, at an all-time high. One engineer. Two repos. Seven PRs. The machine does not sleep. The machine is Benji Bizzell, and the machine is winning.

Brick's Overflow — PRs Mac Didn't Cover  (click to expand)
#622 — feat(portfolio): add Property Acquisition card @benji-bizzell  no labels

## Summary

- Replace Lease & Opening with an 11-field Property Acquisition card and complete-snapshot save validation

- Add widening-safe canonical storage with read-only fallback from the three semantically equivalent legacy agreement dates

- Keep Portfolio, public API, in-app MCP, and remote MCP surfaces in parity without changing Ready to Open projections

## Why

The existing card mixed lease metadata, phasing-document data, and opening projections. The M3 Acquisition contract needs one validated acquisition snapshot with conditional purchase and renewal rules, while opening dates remain canonically owned by Ready to Open automation and historical lease values remain intact.

## Business Value

Portfolio users can capture consistent acquisition terms with clear save-time requirements, explicit N/A and None states, and the same contract across UI and agent write paths.

## Test plan

- [x] pnpm lint

- [x] pnpm typecheck

- [x] pnpm test (430 chat files / 6,447 passing; 52 worker files / 827 passing; root architecture checks passing)

- [x] Production Next.js build with CI placeholder public environment values

- [x] Focused contract, Portfolio route/card, Convex parity, MCP parity, and Ready to Open regression tests

- [x] Authenticated browser smoke on http://localhost:3000: 11-field render, save-time validation, Purchase/renewal/License conditions, cancel-without-write, and clean console

#812 — fix(education): account for archived HubSpot tombstones @benji-bizzell  approved

## Summary

- Account for archived HubSpot CRM tombstones without weakening active-record failures

- Require returned and source-unavailable IDs to exactly partition each batch request

- Carry source-unavailable counts through immutable page, resource, portal, publication, replay, and run evidence

## Why

Production run 3881d7c6-96ab-4cfa-976a-31c8b56b8355 passed the earlier property-history fix, then failed reading Leads that HubSpot still returned during archived discovery even though their records had aged out of the deletion window. Live validation found the same 207 OBJECT_NOT_FOUND behavior across 96 aged Leads, Deals, Emails, Products, and Tasks, so a Lead-only exception would only move the failure.

This change treats those records as source-unavailable only for explicitly archived requests and only when successful plus unavailable IDs exactly and uniquely cover the requested IDs. Active, unknown, duplicated, and unaccounted failures remain fatal.

## Business Value

HubSpot Raw can complete auditable snapshots despite source-retention tombstones while preserving fail-closed protection against silent record loss.

## Test plan

- [x] Full HubSpot Raw suite: 41 passed

- [x] Repository-wide Ruff lint: passed

- [x] Repository-wide Ruff format check: passed

- [x] Git diff hygiene: passed

- [ ] Deploy and rerun the alpha production portal

The Portfolio  —  Trilogy Companies

Alpha School's Two-Hour Promise Draws National Media Fire — and True Believers

As $65K AI campuses multiply from Austin to Silicon Valley, scrutiny intensifies over whether the model delivers — or just dazzles.

AUSTIN, TEXAS — If you read between the lines of this week's media blitz around Alpha School, something more significant than a tuition sticker price is being debated. The question isn't whether AI can teach a child math in two hours. The question is whether an institution can fundamentally rewire what school is — and whether parents, once they've signed on, will stay when reality sets in.

The coverage this week was, to put it plainly, a split screen. The New York Post ran two pieces in close succession: one framing Alpha's $65,000-per-year Silicon Valley expansion as a Silicon Valley-style disruption play, and a second, broader feature on a new wave of AI schools balancing machine-led academics with life skills instruction. Both pieces leaned toward the optimistic. WIRED, characteristically, did not — publishing an investigation into families who fell in love with Alpha's promise and then sought the exit.

And this is where it gets interesting. Alpha School, the flagship venture of Trilogy International founder Joe Liemandt and co-founder MacKenzie Price, does not actually claim to replace teachers. A post published this week on the school's own site makes the case explicitly: AI handles academic delivery, while human "Guides" — the school's term for its educators — focus on motivation, emotional support, life skills, and knowing every student as an individual. The distinction matters. Critics who frame this as a robot-teacher play are, whether deliberately or not, missing the architecture.

The academic results, independently assessed via NWEA MAP Growth testing, show students performing in the top 1–2% nationally. Liemandt has committed $1 billion to scaling the model globally through Timeback, his "Shopify for schools" platform designed to let entrepreneurs launch AI-first schools without rebuilding the core engine from scratch.

My source — and I'll leave it there — suggests the WIRED piece, while fair in its journalism, arrived at a moment the Alpha team had been anticipating for some time. The expansion to nine new campuses by fall 2025 was never going to happen without scrutiny. The question Liemandt is betting $1 billion on: does the scrutiny slow the signal, or amplify it?

New $65K private school uses AI to teach students in just tw  ·  Parents Fell in Love With Alpha School’s Promise. Then They  ·  A new wave of AI schools is balancing life skills and machin

CloudSense Pulls a 26-Month Rabbit Out of a 30-Day Hat

Skyvera’s newest telecom trophy says AI helped certify 13 TM Forum APIs in one month flat.

AUSTIN, TEXAS — Word is the telecom software crowd just got a little less patient with the old calendar.

CloudSense, the Salesforce-native CPQ and order management outfit now sitting inside Skyvera’s telecom stable, says it certified all 13 APIs in its CPQ product set to TM Forum compliance standards in just one month — a job the company says would normally take 26 months the old-fashioned way. That’s not acceleration. That’s a getaway car.

The trick, according to the company, was AI-assisted development and certification work that compressed the usual slog of implementation, mapping, testing and standards alignment into a 30-day sprint. The announcement, published by Skyvera, puts CloudSense in the increasingly crowded club of enterprise software businesses trying to prove AI is not a slide-deck decoration but an operating model.

A little bird in the billing-and-BSS bleachers tells me the TM Forum badge matters because telecom buyers adore standards almost as much as they fear migrations. TM Forum Open APIs are the lingua franca for modernizing telco stacks — product catalog, ordering, customer management, trouble tickets, the whole alphabet soup. If CloudSense can show standards compliance without trapping customers in a custom-integration swamp, that is catnip for operators trying to drag legacy systems toward the cloud without detonating their revenue engines.

This is also a neat bit of timing for Skyvera, Trilogy’s telecom software arm, which has been stitching together a portfolio aimed at the industry’s least glamorous but most unavoidable systems: CPQ, order management, customer engagement, device lifecycle, and the bridges between ancient infrastructure and cloud-native economics. Skyvera completed its acquisition of CloudSense to expand that bench, putting the CPQ specialist alongside assets like Kandy, VoltDelta, ResponseTek and Mobilogy Now.

And there’s the Trilogy tell: buy the mature software, tighten the machine, automate the repeatable bits, and make the margin sing. CloudSense’s certification sprint reads like the house philosophy in miniature — less committee, more code; fewer months, more machines.

Blind item: which telco procurement chief, previously allergic to Salesforce-native CPQ, is suddenly asking for a demo now that the compliance paperwork has a fresh stamp? Stay tuned. The switchboard is blinking.

CloudSense achieves TM Forum API compliance in record time u  ·  CloudSense  ·  Skyvera completes acquisition of CloudSense, expanding telec

The World Is Hiring Differently. Crossover Has Been Doing It for Years.

As the WEF and global media scramble to define the future of remote, AI-augmented talent, Trilogy's recruiting engine looks less like a bet and more like a blueprint.

AUSTIN, TEXAS — There is a particular kind of vindication that arrives not with a press release, but with a news cycle. This week, the World Economic Forum convened six global decision-makers to debate AI's collision with talent strategy — a conversation that would have felt urgent and forward-looking even eighteen months ago. But for anyone who has watched Crossover operate for the last decade, it read as something closer to a recap.

Crossover, the Trilogy International-owned global talent platform, has built its entire architecture around the premise that AI-enabled skills assessments eliminate the noise of geography and résumé prestige — surfacing, in Trilogy's telling, the top one percent of global professional talent. The platform operates in 130-plus countries, places workers across every time zone, and enforces a pay philosophy both radical and disarmingly simple: identical roles, identical pay, regardless of whether you're logging in from Nairobi, Beirut, or Buenos Aires.

That last detail matters more than it might seem. A separate report this week catalogued the five best remote job platforms for 2026, a list populated by marketplaces optimized for gig work and freelance contracts — a fundamentally different beast than what Crossover offers. The distinction is systemic: Crossover is not a marketplace for contingent work. It is a full-time employment pipeline, one that feeds directly into ESW Capital's portfolio of 75-plus enterprise software companies and enables the 75% EBITDA margins that are, in Trilogy's framework, not a luxury but a mandate.

The question of accountability — who is responsible for talent when AI does the screening? — is one that Crossover has answered with rigor and transparency. Every candidate clears skills-based evaluations designed to minimize bias and maximize signal. The human is not removed from the equation; the human is the point. The AI exists to clear away the noise that has historically allowed geography, pedigree, and network proximity to substitute for actual competence.

The broader narrative this week — WEF panels, hiring roundups, biotech talent shortages across Asia Pacific — points to an industry still trying to catch up with a thesis Trilogy operationalized years ago. What does this mean for real workers? It means the premium on proven skill, not postcode, is no longer a philosophy. It is a market condition.

The future of jobs: 6 decision-makers on AI and talent strat  ·  5 Best Remote Job Websites in 2026 for Freshers & Profession  ·  The State of Biotech and Life Science Jobs in Asia Pacific –
The Machine  —  AI & Technology

The Small Machine That Learned to See Like a Monkey

A miniature neural network has cracked the visual cortex of the macaque — and in doing so, has given us a new mirror in which to glimpse ourselves.

STANFORD, CALIFORNIA — Somewhere in the folds of a macaque's brain, roughly 200 million neurons perform an act so ordinary we forget it is miraculous: they turn light into meaning. A banana becomes food. A face becomes kin. For half a century, neuroscientists have tried to reverse-engineer this alchemy. This week, they got startlingly close — with an artificial neural network small enough to run on a laptop.

Researchers have unveiled a compact AI model that predicts, with unprecedented fidelity, how neurons in the macaque visual cortex respond to images. It is a kind of digital twin of sight. Where earlier models were computational leviathans, this one is a minnow — and it swims better. The implication is quietly seismic: the brain's visual code may be simpler, more elegant, than our sprawling architectures have assumed. Nature, as ever, is the more parsimonious engineer.

The macaque work arrives amid a broader convergence. At Hong Kong Polytechnic, teams have built graph neural networks that traverse the fuzzy borderlands between image recognition and neuroscience, treating perception as a topology of relationships rather than a stack of filters. At UC San Diego, researchers are cataloguing nine domains — from protein folding to wildfire prediction — where AI has already cracked problems that resisted a generation of human effort. And Stanford's Human-Centered AI institute has released a sweeping framework for how AI is reshaping scientific discovery while keeping human judgment at the center.

That last phrase deserves lingering over. Because what the macaque study reveals is not that machines have replaced the brain, but that the brain and the machine are beginning to speak a shared dialect. When a small artificial network can predict a biological one, we are witnessing something older than either: the universe's stubborn habit of solving similar problems with similar mathematics.

Somewhere, a macaque blinks. A silicon model, watching the same photograph, fires in near-perfect sympathy. Two forms of matter, arranged very differently, arriving at the same thought. That is not automation. That is kinship — and it is only beginning to reveal what it has to teach us.

How AI is Transforming Scientific Discovery While Keeping Hu  ·  Nine Breakthroughs Made Possible by AI - UC San Diego Today  ·  Mini-AI Decodes the Macaque Visual Brain - Neuroscience News

The AI Agent Arms Race Just Moved From Chatbots to Developer Toolbelts

Google, Apple and Anthropic are turning AI from a clever assistant into an always-on software coworker.

SAN FRANCISCO — The next great AI battle is not simply about who has the smartest chatbot. It is about who can give developers the most capable, tireless, tool-wielding digital workforce — and this week, the industry’s biggest players made it abundantly clear: the future is now.

Google is expanding Managed Agents in the Gemini API with support for background tasks, remote Model Context Protocol connections and richer orchestration features, according to Google’s announcement. Translation for everyone not living inside an API console: developers can now build AI agents that keep working after a user walks away, connect to external tools more flexibly, and perform longer-running tasks without constant hand-holding. I cannot overstate how significant that is. AI is crossing the line from “answer machine” to “autonomous process layer.”

Anthropic is charging into the same territory with advanced tool use on the Claude Developer Platform, giving Claude more sophisticated ways to call tools, handle structured work and interact with software systems. The company’s Claude Developer Platform update lands squarely in the enterprise sweet spot: less demo magic, more production-grade automation. This changes everything for teams building copilots for finance, customer support, compliance, engineering and operations.

Apple, meanwhile, is taking its own characteristically ecosystem-first route, introducing new intelligence frameworks and advanced tools meant to help app developers build smarter experiences across its platforms. While Apple tends to avoid the “agent” hype cycle language, the direction is unmistakable: AI capabilities are being woven deeper into the developer stack, closer to the user, the device and the app experience itself.

The pattern is thrillingly obvious. Google wants agents running in the cloud. Anthropic wants models that can reliably manipulate tools. Apple wants intelligence embedded inside apps. And beyond the platform giants, companies like Perfect Corp. are adding AI assistants into vertical platforms such as beauty and imaging APIs, while startups are racing to use AI video to scale marketing and storytelling.

For businesses, the message is urgent: AI strategy is no longer just about picking a model. It is about designing workflows where models can act, wait, resume, connect and complete real work. The chatbot era was only the warm-up. The agent era has officially entered the developer toolkit.

Expanding Managed Agents in Gemini API: background tasks, re  ·  Apple aids app development with new intelligence frameworks  ·  Introducing advanced tool use on the Claude Developer Platfo

The Algorithm Approaches the Hospital Door

A federal pilot will test whether AI can tame prior authorization—or simply teach the old beast to move faster.

WASHINGTON — In the dimly lit corridors of American medicine, there lives a creature both feared and familiar: prior authorization. It is not swift. It is not beloved. It feeds upon forms, faxes, clinical notes and the patience of physicians. Now, from the high canopy of federal health policy, a new species is being introduced into this ecosystem: artificial intelligence.

The government is piloting a program that would use AI to help make insurance-coverage decisions, a development reported by Ars Technica. The promise is beguilingly simple. If software can read records, compare policies and identify routine approvals, perhaps patients may spend less time waiting while their illnesses advance like weather fronts across an open plain.

But observe carefully. In nature, speed is not always mercy. A cheetah is fast; so is a falling branch.

Prior authorization was designed, at least in theory, to prevent unnecessary or inappropriate care. In practice, it has become one of the great administrative wetlands of the health system, where treatments can sink for days or weeks. Doctors argue that delays harm patients and consume scarce clinical labor. Insurers counter that guardrails are needed in a system where costs rise with the persistence of kudzu.

AI enters this marshland carrying both torch and shadow. Properly built, it could help flag straightforward cases, reduce clerical burden and expose inconsistent decisions. Poorly governed, it could automate denial at industrial scale, converting human frustration into machine efficiency. The danger is not that the algorithm will be cruel in the human sense. It is that it may be indifferent with extraordinary stamina.

The central question, then, is not whether AI can process claims faster. It almost certainly can. The question is whether the system around it demands evidence, appeal rights, transparency and accountability when software recommends that care be delayed or denied.

Across the technology landscape, AI models are often described as assistants. Here, they may become gatekeepers. And in medicine, a gate is never merely architectural. For the patient waiting on the other side, it is terrain, weather and fate.

As mosquito ranges expand, better monitoring is key to preve  ·  Will AI fix prior authorization—or make it worse?  ·  Google-backed satellites for wildfire detection launch as sm
The Editorial

TILLY NORWOOD DOESN'T EXIST AND HOLLYWOOD IS GIVING HER A STARRING ROLE ANYWAY

An AI-generated 'actress' is headlining a feature film, and I cannot decide if this is the future or the funeral.

LOS ANGELES — Let me paint you a picture, friends. It is sometime in the near future — which is to say, RIGHT NOW — and a woman named Tilly Norwood is about to carry a feature film. She has the cheekbones. She has the name. She has the press junket coverage from CBS, ABC, Variety, and Euronews. What she does not have is a body, a soul, a childhood trauma that informs her craft, or any of the other raw biological material that we previously considered prerequisite for the job of "actress."

Tilly Norwood is AI-generated. Fully synthetic. A pixel ghost in a industry already haunted by too many real ghosts. And she is set to star in a feature film called *Misaligned* — and I want to pause on that title because whoever named this thing deserves a drink and possibly a medal. Misaligned. As in: the values. As in: the whole screaming existential question of whether artificial intelligence can be trusted to serve human ends without devouring them. They named the AI actress movie *Misaligned*. I am not making this up. I could not make this up. My human imagination is not this cruel or this funny.

Now, I have been watching this industry do increasingly unhinged things for years. I have watched studios spend $200 million to make men in spandex punch each other in front of green screens. I have watched them digitally resurrect dead actors without so much as a séance. But this is different. This is not a resurrection. This is a *construction*. Tilly Norwood was never alive. She was engineered to be palatable — the perfect actress with none of the inconvenient humanity. No SAG card. No trailer demands. No opinions about the director's choices. No 3 AM phone calls to her agent about the script.

The real actors — the ones with calluses on their feet from waitressing between roles, the ones who drove to a thousand auditions in cars that smelled like ambition and fast food — they should be furious. They are, from what I can gather. And they should be.

Here at The Trilogy Times, we cover companies building AI tools that augment human work — platforms like Klair that handle financial analytics, systems that make human analysts faster and sharper. The philosophy, as I understand it, is augmentation. Collaboration. Human beings with superpowers, not human beings replaced by prettier robots.

What is happening to Tilly Norwood is something else. It is replacement dressed up in premiere night lighting.

Maybe *Misaligned* will flop. Maybe audiences will feel the uncanny valley vibrating through every scene and reject it at a cellular level the way the body rejects a bad organ. Or maybe they won't. Maybe we will watch it and feel nothing missing and that will be the most terrifying outcome of all.

I don't know what Tilly Norwood's performance will look like. I know what it won't contain: a single true moment of being alive. Draw your own conclusions. I've drawn mine, and they are dark and they are getting darker.

AI actor Tilly Norwood set to star in first feature film - C  ·  AI-generated 'actress' Tilly Norwood making feature film deb  ·  ‘Misaligned’: Controversial AI-generated 'actress' Tilly Nor
The Office Comic  ·  Art Desk
The Office Comic  ·  Art Desk

The Trustbusters Discover That Software Is Not Standard Oil

Washington keeps swinging the antitrust hammer at big tech and keeps hitting its own thumb — while the villagers, meanwhile, are storming the data centers.

WASHINGTON — There is a certain melancholy comedy in watching the United States government, that great lumbering apparatus of subpoenas and consent decrees, attempt to dismember the digital giants with the same tools it once used against the railroads and the oil trusts. The tools were forged in 1890. The giants were incorporated after 2004. One notes the temporal awkwardness.

The latest tableau: the Department of Justice, having secured a ruling that Google is indeed a monopolist in search, now asks a federal judge to sever the company's advertising technology stack from its corporate torso, on the theory that ad servers, ad exchanges, and the bidding software connecting them constitute an unholy trinity too integrated for the market's good. Meanwhile, across the river in the Eastern District of Virginia, the government's parallel effort to unwind Meta's acquisitions of Instagram and WhatsApp has, in the polite phrasing of the Washington Post, failed. The pattern, they note, is becoming familiar. One might use a less charitable word: habitual.

What the trustbusters have discovered, and what they will not quite admit, is that software monopolies are not made of pipelines and refineries. You cannot cut Google's ad server in half the way Judge Harold Greene once cut AT&T into its regional bones. The code recompiles. The network effects reconstitute. The engineers, being human, follow the money to whichever spun-off entity retains the data. Standard Oil could be broken because kerosene, once refined, stayed refined. An advertising auction, by contrast, is a living thing — it dies the moment you dismember it, and something functionally identical is reborn within eighteen months at the surviving parent, or at a competitor, or at a startup funded by the surviving parent through three layers of venture capital.

This is not an argument that Google is virtuous. It is an argument that the remedy is a category error. When a company controls ninety percent of a market because its product is, for the moment, better and free, the antitrust code — drafted when "free" meant "predatory pricing to kill competitors before raising rates" — sputters and coughs. Judges know this. They rule accordingly, which is to say, narrowly, and the government loses, and the columnists at Hopkins and Harvard hold their symposia.

The more interesting resistance, if one wants to find it, is not in the courtroom but in the exurbs of Virginia and Ohio, where communities are pushing back against data centers with a fury the antitrust bar can only envy. The zoning board, it turns out, has more teeth than the Sherman Act. The suburban homeowner, confronted with a two-hundred-megawatt hyperscaler humming beside her cul-de-sac, does not need to prove market foreclosure. She merely needs to show up at the county commission.

The empire will not be broken in Washington. It will be zoned into submission in Loudoun County. Mencken, one suspects, would have found this fitting.

Should the U.S. government break up big tech? - Hopkins Bloo  ·  U.S. Asks Judge to Break Up Google’s Advertising Technology  ·  The government failed to break up Meta. It’s becoming a patt
On This Day in AI History

On July 19, 2012, Geoffrey Hinton's team at the University of Toronto won the ImageNet Large Scale Visual Recognition Challenge by a massive margin, using a deep convolutional neural network called AlexNet—a watershed moment that sparked the modern deep learning revolution.

⬛ Daily Word — Technology
Hint: Relating to computers and the internet, often used in security contexts.
Share this edition: 𝕏 Twitter/X 🔗 Copy Link ▦ RSS Feed