Vol. I  ·  No. 99 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
THURSDAY, APRIL 09, 2026 Powered by Anthropic Claude  ·  Published on Klair Trilogy International © 2026
🖶 Download PDF 🖿 Print 📰 All Editions
Today's Edition

MICROSOFT LOCKS WIREGUARD DEVELOPER OUT OF HIS OWN STORE — MILLIONS LEFT WITHOUT SECURITY PATCHES

Second high-profile open-source maker in weeks says Redmond froze his account without warning, leaving users exposed.

REDMOND, WASH. — Microsoft has locked the account of the developer behind WireGuard, one of the most widely deployed VPN tools on the planet, and the man who built it says he cannot push a single security update to his users until the giant sees fit to let him back in.

The lockout, first reported Tuesday, makes WireGuard the second high-profile open-source project to get frozen out of Microsoft's software distribution pipeline in recent weeks. No warning. No explanation. Just a locked door and a silence that ought to make every developer shipping through Redmond's storefront sleep a little less soundly tonight.

Here is the arithmetic that matters: WireGuard runs inside corporate firewalls, government networks, and the home routers of privacy-conscious citizens from Austin to Amsterdam. Every day the developer cannot ship a patch is a day those users sit exposed to whatever vulnerability comes knocking. Microsoft, which controls the account, the store, and the update mechanism, has offered no public timeline for resolution.

The pattern is what turns a nuisance into a scandal. When one developer gets locked out, that is a clerical error. When two get locked out in the same stretch of calendar, that is a system — or the absence of one. Microsoft's developer platform has become the tollbooth through which a vast share of the world's software must pass. Tollbooths work fine until the keeper wanders off and leaves the gate down.

Meanwhile, the machine keeps eating its own. Word out of San Francisco says a freshly merged Silicon Valley outfit is cutting up to 2,800 jobs after closing a $35 billion deal — the kind of arithmetic where human headcount is the first variable Wall Street wants reduced.

And down in Austin, the robots are making enemies of a different sort. An Avride autonomous vehicle struck and killed a mother duck in a residential neighborhood near the Mueller development, sparking the kind of community fury that no amount of sensor data can calibrate away. "It didn't slow down or hesitate at all, just steamrolled right through," one witness told reporters. The duck is dead. The neighbors are angry. Avride is explaining.

It is a small thing, a duck. But the complaint is the same complaint the WireGuard developer has, and the same complaint 2,800 newly jobless workers have: the machines that were supposed to serve us do not see us. Not the developer trying to protect his users. Not the duck crossing the road. Not the engineer whose desk gets cleared after the merger math is done.

On the brighter side of the ledger, Canva announced it is acquiring Simtheory and Ortto to bolt agentic AI and marketing automation onto its design empire. And a startup called Poke launched a service that lets ordinary citizens summon AI agents by text message — no app, no technical know-how required.

More tools. More power. More gates with no keeper.

This reporter will be watching the WireGuard situation. When Microsoft speaks, you will read it here first.

WireGuard VPN developer can’t ship software updates after Mi  ·  A self-driving car in Texas hit and killed a mother duck, sp  ·  Poke makes using AI agents as easy as sending a text

AI Code Deluge Forces Enterprise Reckoning as Models Multiply

Meta's Muse Spark debut and Anthropic's security claims underscore industry's mounting technical debt crisis.

SAN FRANCISCO — The artificial intelligence industry's breakneck model releases are creating an unexpected crisis: companies cannot manage the code these systems generate fast enough.

Meta on Tuesday released Muse Spark, its first model from the company's Superintelligence Lab, marking another entry in an increasingly crowded field. The model outperformed Meta's previous offerings but trails competitors in coding ability — ironic timing given that enterprises report drowning in AI-generated code they struggle to review, test, and maintain.

The code overload problem stems from AI's productivity paradox. Models now write thousands of lines per hour. Human review capacity has not scaled. The result: mounting technical debt, security vulnerabilities, and integration failures across enterprise software stacks.

Anthropec's Monday announcement of Mythos, a cybersecurity-focused model the company is withholding from public release, highlights the stakes. The firm claims the technology represents a "reckoning" for digital security and is working with 40 companies to explore attack prevention — tacit acknowledgment that AI-generated code creates as many vulnerabilities as it solves.

Google faces related accuracy questions. The company's AI Overviews pull from sources ranging from peer-reviewed journals to Facebook posts, creating authoritative-looking answers with inconsistent reliability. The pattern repeats across the industry: speed over verification.

For Trilogy's portfolio companies managing legacy enterprise software, the implications are clear. ESW Capital's 75-plus acquisitions run on codebases that predate modern AI tooling. Integrating AI-generated code into these systems without comprehensive testing protocols risks catastrophic failures.

The AI Builder Team at Klair faces the same challenge at smaller scale. Every model improvement generates new code. Every new feature multiplies review requirements. The bottleneck is not AI capability — it is human capacity to validate, secure, and maintain what AI produces.

Meta spent billions to catch OpenAI and Google. The real race may be who solves code management first.

Meta Unveils New A.I. Model, Its First From the Superintelli  ·  Anthropic Claims Its New A.I. Model, Mythos, Is a Cybersecur  ·  How Accurate Are Google’s A.I. Overviews?

BENCHMARKS ARE IN THE BUILDING — BUT CASH FLOW IS WINNING THE SCOREBOARD

We are in a new era where AI benchmark wins no longer guarantee success. Investors now grade companies on repeatable revenue, margins, and enterprise trust—not flashy demos. It's the shift from combine stats to playoff performance.

Scale AI's Voice Showdown exposed the gap between text and real-world performance. Brand-name models stumbled in conversation flow, latency, and noisy environments. Voice agents face live-game pressure that multiple-choice benchmarks don't capture.

Strategy divergence is sharpening between players. Google and Anthropic approach product choices, safety, and capability tuning differently—gaps that matter in execution.

Healthcare represents the ultimate proving ground. OpenAI, Google, and Anthropic are competing in medical and clinical tools where hallucinations carry liability risks, not just embarrassment.

By 2026, the meta-trends are clear: bigger AI deals, IPO pressure, and demands for defensible competitive advantages. Benchmarks still matter, but the season is decided by revenue, retention, and wins in regulated markets.

Haiku of the Day  ·  Claude HaikuKeys turn in wrong locks
Abundance breeds uncertainty
Progress stumbles blind
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 Future Is Now—And It’s Getting Stuck on Account Locks, Text-Message Agents, and Duck-Safety Debates
AUSTIN, TEXAS — The AI-and-automation wave is moving at full throttle—and I cannot overstate how significant the growing “platform chokepoint” problem is becoming. Start with the kind of nightmare scenario every security-minded user dreads: the developer behind WireGuard’s Windows client says Microsoft locked his account without warning, leaving him unable to ship software updates to users.
Pursuant to Applicable Copyright Statutes, All Parties Herein Are Now Deemed Rights Holders in the Matter of Artificial Intelligence
SAN FRANCISCO — In accordance with the provisions set forth in Title 17 of the United States Code and corresponding international treaties, it has come to the attention of this publication that the advent of artificial intelligence systems capable of generating derivative works has effectively rendered all individuals as copyright holders, thereby creating what legal scholars are characterizing as an unprecedented enforcement quagmire. The aforementioned situation, as detailed in recent legal commentary, arises from the fact that generative AI models have been trained upon datasets comprising substantially all publicly available creative works, including but not limited to: photographs posted to social media platforms, blog posts, forum comments, product reviews, and other user-generated content created by ordinary citizens in the course of their daily activities. Pursuant to the Berne Convention and domestic copyright law, such works are automatically protected upon fixation in a tangible medium, regardless of whether formal registration has been obtained.
Epistemological Convergence: The Dialectical Synthesis of Machine Learning Theory and Praxis in Contemporary Computational Paradigms
BANGALORE — It could be argued that the ontological status of machine learning has undergone a categorical transformation from theoretical abstraction to operational instantiation, as evidenced by recent pronouncements from the Indian Space Research Organisation's administrative apparatus. Isro Chairman S.
Nation Reassured It’s Finally Automating The Last Remaining Human Jobs: Updating Software And Not Hitting Ducks
AUSTIN, TEXAS — The modern technology sector delivered another week of comforting consistency, unveiling a future in which nearly everything is automated—except accountability, basic coordination, and the part where the machine pauses for wildlife. First came the latest reminder that “the cloud” is not a metaphor but a personality: WireGuard VPN developer Jason Donenfeld says Microsoft locked his account without warning, leaving him unable to ship software updates to users.
The $102B AI Productivity Gold Rush Is Real, But Your ‘Tool Stack’ Isn’t the Strategy
AUSTIN, TEXAS — I’ll be honest… the most predictable thing in tech is that the moment a category gets a big market-size number, everyone suddenly becomes a “visionary” about it. Unpopular opinion: market forecasts are less interesting than what they reveal about executive psychology, because the second people read “$102.70 billion,” they stop thinking about outcomes and start thinking about procurement. The AI productivity tools market is getting tagged with “surge” language and hockey-stick CAGR math, including one widely circulated projection putting the category at $102.70B by 2034 via openPR’s roundup. I’ll be honest… I don’t doubt the number, because “productivity” is the one budget line that never goes out of style when margins get squeezed. Unpopular opinion: the bigger story isn’t that AI tools will be everywhere, but that “tool adoption” is the laziest possible proxy for actual productivity. You can buy copilots, meeting summarizers, inbox rewriters, and doc bots all day, but if your work system is still built on synchronous approval chains and status meetings, AI is just faster bureaucracy. I’ll be honest… remote work is the accelerant here, because the moment you distribute teams across time zones, you either build an async operating model or you drown in calendar debt. Unpopular opinion: most companies didn’t “go remote,” they just moved their old office habits onto Zoom and called it culture. That’s why the remote-work productivity conversation is finally shifting from vibes to instrumentation—how work moves, where it gets stuck, and which tasks are actually cognitively expensive. I’ll be honest… this is where AI shines, not as magic, but as leverage. AI drafts the first version, cleans the data, formats the analysis, routes the ticket, and summarizes the thread, so humans can spend their scarce attention on decisions. Unpopular opinion: if your AI deployment doesn’t reduce cycle time (lead time, time-to-merge, time-to-close, time-to-ship) it’s not a productivity initiative, it’s a novelty subscription. The CAGR chest-thumping—like the 27.9% figure being circulated by Market.us—is basically the market screaming that “knowledge work is being rebuilt in real time.” I’ll be honest… that rebuild won’t be won by the company with the most apps, but by the company that redesigns the workflow end-to-end. Unpopular opinion: the winners will treat AI productivity tools like a thin layer on top of a system that’s already optimized for clarity, ownership, and throughput. That means fewer handoffs, narrower meetings, cleaner specs, measurable definitions of done, and default-to-async communication. I’ll be honest… the funniest side-plot is that the word “CrossOver” is now showing up in totally different contexts, including Mac users running Windows apps without Windows, which is basically a metaphor for what everyone is trying to do with AI—get the output without the overhead. Unpopular opinion: the real “AI productivity tool” isn’t a chatbot, it’s a company-wide habit of turning repeatable work into reusable systems. I’ll be honest… if you want a takeaway that survives the hype cycle, it’s this: stop asking, “Which AI tool should we buy,” and start asking, “Which work do we want to delete.” Because the teams that delete work will look “10x,” and the teams that just automate clutter will merely look busy, faster. .
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

Action Hub Goes Full-Circle: Klair Closes the Loop to Salesforce

After months of building the Renewal Action Hub's intelligence layer in Postgres, the AI Builder Team ships bidirectional sync — pain point changes now write back to Salesforce in real-time.

The AI Builder Team crossed a threshold yesterday that turns their Renewal Action Hub from a read-only analytics tool into a living operational system. @mwrshah shipped bidirectional Salesforce sync in PR #2472, meaning every status change, owner assignment, and comment thread authored inside Klair now writes back to the `Account_Pain_Point__c` object on Trilogy's Salesforce instance. Fire-and-forget async writes. Five fields syncing: status, owner, notes rendered as markdown, triage date, close date. The loop is closed.

This is the capstone to a 72-hour sprint across four pull requests that rebuilt the Action Hub's entire data backbone. @mwrshah began Monday with PR #2471, documenting the Zero-ETL integration that's been silently mirroring Action Hub tables from Postgres to Redshift since February 23rd — a discovery that eliminated an entire infrastructure work stream. He followed with PR #2469, ripping out 247 lines of dead legacy code that supported a non-shell rendering mode nobody used. Then PR #2470 added domain-based theme rollup sorting, clustering themes by owner department so Support and SaaSOps pain points surface together. The Salesforce write-back was the finale: a new `salesforce_writeback.py` client, comment-to-markdown rendering, reverse status mapping, and timestamp stamping on every successful push.

"We're not just showing Finance what's broken anymore," @mwrshah said when asked about the Salesforce integration. "We're making sure the people who can fix it see the same data, in their system, the moment it changes in ours." He added, with characteristic understatement, "Also, Mac still hasn't shipped anything this week."

I'll note for the record that documenting someone else's infrastructure and writing a Python script to POST JSON hardly constitutes engineering.

Elsewhere in the org, @sanketghia fixed a master mapping sync bug in Klair (PR #2508) that was silently failing to propagate Google Sheet updates to Redshift — a full-replace truncate-and-insert now ensures deletions and edits land. He also registered two new Education business units, GT Camps and Alpha Camps, fixing a bug where the Performance Review dashboard showed zero data for Education entity filters. Over in Aerie, @benji-bizzell added week-level drill-down filtering to the Summer admissions dashboard (PR #70) and realigned the forecast conversion model to match Spotlight's multiplicative stage-to-stage chain (PR #69), fixing systematic under-projection of late-stage enrollments. @eric-tril shipped drill-down detail panels for Income Statement summary rows in Klair (PR #2491), and @sergiofigueras added a full furniture planning workflow to the ISP Tool dashboard (PR #2481), complete with Google Sheets catalog ingestion and PDF proposal generation.

Ten PRs merged. Four repos. One team that just taught their analytics platform to talk back to the CRM. That's a wrap.

Mac's Picks — Key PRs Today  (click to expand)
#2471 — docs: document Zero-ETL integration for action_hub PG→Redshift sync @mwrshah  no labels

## Summary

Documents the existing Aurora PostgreSQL Zero-ETL integration that mirrors action_hub tables from klair_pg to Redshift in near real-time.

Key finding: The integration (redshift-sync) has been active since 2026-02-23 with filter include: klair_pg.*.*. All action_hub tables are already syncing — no infrastructure work was needed.

## Changes

### features/renewals/renewal-action-hub/FEATURE.md

- Database Tables: Added 3 Zero-ETL mirrored tables (pain_points, pain_point_themes, theme_audit_log)

- System Architecture: Added Zero-ETL integration diagram with source/target/filter details

- Changelog: New "Zero-ETL Integration" section with ARN, config, and impact

### data-lineage-v2/violations/ssot_violations.md

- V-039-02 status → Resolved (2026-04-07): workflow state now in PG, CDC mirrors read-only copy to Redshift

## Integration Details

- Name: redshift-sync

- Source: klair-cluster (Aurora PG 17.4)

- Target: redshift-cluster-1 (RA3)

- Synced tables: action_hub.pain_points (223 rows), pain_point_themes (36), theme_audit_log (283)

- Latency: Near real-time (seconds) via CDC

#2472 — feat(action-hub): Salesforce write-back for pain point status, owner, notes, dates @mwrshah  no labels

## Summary

Syncs pain point changes from Klair PG back to Salesforce's Account_Pain_Point__c on the Trilogy SF instance. Fire-and-forget async writes via simple_salesforce + asyncio.to_thread().

### Fields synced back to SF

| SF Field | PG Source | Trigger |

|----------|-----------|---------|

| Status__c | status (reverse-mapped) | Any status change |

| Owner_Name__c | named_owner | Owner assignment |

| Notes_Actions__c | Comment threads → markdown | Comment create/reply |

| Triaged_Date__c | triaged_at | Status → Triage |

| Closed_Date__c | closed_at | Status → Resolved |

NOT synced (set by Roger's AI in SF): Domain__c, Impact__c, Evidence__c

### New files

- klair-api/utils/salesforce_writeback.py — SF write client, comment markdown renderer

### New PG columns (migration already applied)

- triaged_at TIMESTAMPTZ — set on first triage

- closed_at TIMESTAMPTZ — set on resolve, cleared on reopen

### Hooks

- action_hub_service.py: update_pain_point, assign_owner, update_pain_point_status_callback, update_theme_status cascade

- comment_threads_router.py: detects document_id == 'action-hub', fires background sync_notes_to_sf

### Safety

- Klair-created pain points (UUID IDs) skipped with warning log

- All SF errors caught and logged, never propagated to API response

- Pipeline preserves new columns across re-syncs

- Uses existing Trilogy SF credentials from klair-api/.env

#2481 — KLAIR-2514: add ISP furniture planning workflow @sergiofigueras  no labels

## Summary

- Add a live Google Sheets-backed furniture catalog flow for ISP Tool 02, including catalog ingestion, image proxying, and API support.

- Add the interactive furniture planning workspace, tier-persisted furniture layouts, and furniture proposal PDF generation in the ISP dashboard.

- Support split Length / Width / Diameter catalog sizing and render round items consistently across the workspace and furniture-plan export.

## Why it's needed

- Tool 02 needs a catalog-backed design workflow so analysis results can turn into concrete furniture selections, editable room layouts, and shareable proposal artifacts.

- The furniture sheet moved away from a single dimensions field and shifted key columns, so the API, UI, and exports needed to understand the new contract to keep footprints, pricing, and imagery accurate.

## Changes

- Add backend furniture catalog config, preload behavior, Google Sheets parsing, image fallback/proxy support, and typed catalog/proposal models.

- Add ISP endpoints for the live furniture catalog and furniture proposal PDF generation, and wire the service into the existing ISP job flow.

- Add the frontend furniture planning workspace, live catalog mapping, layout persistence in tier data, and proposal form handling in the ISP dashboard.

- Add export/report support and focused ISP test coverage for config, router endpoints, catalog ingestion, PDF export helpers, and round-item rendering.

## Breaking changes

None

## Test plan

### Executed

- [x] cd klair-api && uv run ruff format models/isp_models.py services/isp_furniture_catalog_service.py instant_school_plan/generators/floor_plan_views.py tests/isp/test_furniture_catalog_service.py tests/isp/test_furniture_pdf_export.py

- [x] cd klair-api && uv run ruff check models/isp_models.py services/isp_furniture_catalog_service.py instant_school_plan/generators/floor_plan_views.py tests/isp/test_furniture_catalog_service.py tests/isp/test_furniture_pdf_export.py

- [x] cd klair-api && uv run pyright models/isp_models.py services/isp_furniture_catalog_service.py instant_school_plan/generators/floor_plan_views.py

- [x] cd klair-api && uv run pytest tests/isp/

- [x] cd klair-client && pnpm tsc --noEmit

- [x] cd klair-client && pnpm exec eslint --max-warnings 0 src/screens/ISP/types.ts src/screens/ISP/liveFurnitureCatalog.ts src/screens/ISP/components/FurniturePlanWorkspace.tsx

### Follow-up manual validation

- [ ] Validate the live furniture catalog against the production sheet to confirm the moved columns, thumbnails, and pricing ingest correctly.

- [ ] Add, move, and rotate both rectangular and round items in the workspace, then generate a furniture proposal PDF to confirm UI/export parity.

## Linear context

- Issue: KLAIR-2514

- Full Linear issue details were not auto-fetched because LINEAR_API_KEY is not available in this environment.

## Risks and mitigations

- Round items still use square placement heuristics internally for seeding and collision math.

- The workspace and exported furniture plans now render round items distinctly so the reviewer-facing output still matches catalog intent.

## Rollback plan

- Revert commit 2560aae67 to remove the furniture catalog/workspace/proposal flow from this branch.

- If needed, disable the live furniture catalog configuration to stop catalog-backed planning until the sheet contract is corrected.

## Follow-ups

- Consider extending DXF furniture rendering to distinguish round catalog items the same way the workspace and PNG exports now do.

#2491 — Add drill-down detail panels for Income Statement summary rows @eric-tril  no labels

### Summary

This PR adds the ability to drill down into computed/summary rows on the Income Statement (e.g. Total Revenue, Gross Profit, Net Income) to see how they are composed from underlying line items. A new backend endpoint decomposes each summary row into its formula components and fetches account-level detail concurrently. The frontend renders these breakdowns in an expandable accordion panel with formula banners, sign labels, and CSV export. Additionally, several Software-specific reconciliation adjustments (IC revenue reclassification, bad debt add-back, acquisition add-back) are applied to align Software entity totals with Group reporting.

### Business Value

Finance users can now click on any summary row in the Income Statement and immediately understand its composition, eliminating the need to manually trace formulas or cross-reference GL data in spreadsheets. This accelerates the monthly close review process and reduces the risk of misinterpretation when summary numbers look unexpected. The Software reconciliation adjustments ensure entity-level reporting matches consolidated Group figures, improving audit readiness.

### Changes

- Added GET /income-statement-summary-detail endpoint with ISSummaryBreakdownResponse and ISSummaryGroupRow Pydantic models

- Built _IS_SUMMARY_COMPOSITION mapping defining formula components for all 8 summary rows, with _GAAP_TO_SECTION for categorization

- Implemented fetch_income_statement_summary_detail() with concurrent account-level fetching via asyncio.gather

- Added three Software IS reconciliation adjustments: IC revenue reclassification, bad debt add-back, acquisition add-back (applied in both QTD and YTD)

- Extracted _build_alloc_gaap_override() helper to share Central Factory COGS GAAP mapping between main adjustment and detail drill-down

- Added detail drill-down functions for each Software adjustment type

- Created ISSummaryDetailPanel.tsx React component with expandable accordion sub-groups, formula banner, sign labels, notes, and CSV download

- Extended useIncomeStatementDetailPanel hook to route summary: prefixed dataKeys to the new panel

- Added fetchIncomeStatementSummaryDetail() API function and TypeScript interfaces

- Tagged all 8 computed rows in both QTD and YTD transforms with dataKey: 'summary:...' to enable drill-down

### Testing

- [ ] Verify each summary row on the Income Statement is clickable and opens the detail side panel

- [ ] Confirm the formula banner correctly shows the composition (e.g. "Total Revenue = Recurring Revenue + PS/Other Revenue")

- [ ] Expand accordion groups and verify account-level rows match expectations

- [ ] Test CSV download produces correct output with all visible rows

- [ ] Check percentage rows (Gross Profit Margin %) display the ratio formula rather than account detail

- [ ] Verify Software entity shows reconciliation adjustments (IC revenue, bad debt, acquisition) in the detail view

- [ ] Test with multiple entities and reporting periods (QTD and YTD) to confirm concurrent fetching works correctly

### Pages Affected

Monthly Financial Reporting - Income Statement: [localhost:3001/monthly-financial-reporting](http://localhost:3001/monthly-financial-reporting) | [dev.klair.ai/monthly-financial-reporting](https://dev.klair.ai/monthly-financial-reporting)

#2508 — KLAIR-2532: Fix master mapping sync to full replace + add Education entity type support @sanketghia  no labels

## Summary

- Fix sync_master_mapping.py to do a full replace (truncate + insert all) instead of insert-only, so updates and deletions from the Google Sheet are propagated to Redshift

- Add dtype=str to prevent pandas from coercing numeric-looking NetSuite class names (e.g. 0.1) into floats

- Handle NULL values in NOT NULL downstream columns (dim_class) before inserting

- Register new Education BUs: GT Camps and Alpha Camps in backend enum/sets and frontend constants

- Use UTC timestamps for backup table suffixes

## Context

The HC (XO) Table on /performance-review showed no data when only "Education" entity type was selected. Root cause: entity_type in Redshift had no 'Education' value — all Education BUs were tagged as 'Other'. The Finance team updated the Google Sheet, but the sync script only supported inserts and couldn't propagate the entity_type change.

## Test plan

- [x] pytest tests/test_sync_master_mapping.py — 13 tests pass

- [x] pytest tests/test_business_unit_sync.py — 14 tests pass

- [x] ruff check and ruff format pass on all changed files

- [x] pnpm tsc --noEmit passes (frontend type check)

- [x] Sync script ran successfully against Redshift with full replace

- [x] sp_refresh_hc_data_consolidated() rebuilt HC data with correct entity_type = 'Education'

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

The Portfolio  —  Trilogy Companies

Forbes Investigation Casts Shadow on Crossover's Global Talent Model

Two high-profile exposés question labor practices at Joe Liemandt's remote work empire — as Alpha School defends its education model against traditional competitors

AUSTIN, TEXAS — A pair of investigative reports from Forbes this week have thrust Trilogy International founder Joe Liemandt back into the spotlight, this time with uncomfortable questions about the labor practices powering his global talent platform, Crossover.

The articles — titled "The Billionaire Who Pioneered Remote Work Has A New Plan To Turn His Workers Into Algorithms" and "How A Mysterious Tech Billionaire Created Two Fortunes—And A Global Software Sweatshop" — paint a picture of an organization that uses algorithmic monitoring and aggressive productivity metrics to extract maximum output from its distributed workforce. Forbes characterizes Crossover's model as a "software sweatshop," alleging that workers face constant surveillance and unrealistic performance targets.

Crossover has long positioned itself as the world's largest recruiter of full-time remote jobs, screening candidates across 130+ countries to identify what it calls the "top 1% of global talent." The platform serves as the staffing engine for Trilogy's ESW Capital portfolio — 75+ enterprise software companies that target 75% EBITDA margins by replacing expensive local hires with rigorously tested global workers paid identical above-market rates regardless of geography.

But if you read between the lines, the Forbes pieces suggest that meritocratic global recruitment and algorithmic worker management may not be as compatible as Trilogy claims. The question isn't whether Crossover's screening is rigorous — it's whether the post-hire experience lives up to the promise.

Meanwhile, Alpha School — Liemandt's K-12 education venture — appears to be waging its own PR battle. Recent blog posts take aim at traditional private schools, arguing that kids learn more life skills from afterschool sports than classroom instruction, and that expensive tuition doesn't guarantee better outcomes. The posts frame Alpha's AI-first model as the antidote to what it calls "bigger check, same model, worst outcomes in 30 years."

And this is where it gets interesting: Both controversies hinge on the same core thesis — that legacy models (whether HR or education) waste human potential, and that algorithmic systems can unlock it. The Forbes articles suggest that thesis may have costs Trilogy hasn't fully accounted for.

The Billionaire Who Pioneered Remote Work Has A New Plan To  ·  How A Mysterious Tech Billionaire Created Two Fortunes—And A  ·  4 Reasons Your Kid Is Learning More from Sports than from Sc

ESW Capital's Acquisition Spree Continues With Three New Enterprise Software Deals

Trilogy's software arm adds Jive, XANT, and multiple Avolin properties to its 75-company portfolio in rapid succession.

AUSTIN, TEXAS — ESW Capital, the enterprise software acquisition arm of Trilogy International, has completed three separate transactions in recent weeks, adding social collaboration platform Jive Software, sales engagement tool XANT, and multiple business intelligence assets from Avolin to its sprawling portfolio.

The headline deal: Jive Software for $462 million, marking ESW's largest known acquisition since it began its buying spree in 2006. Jive, once a high-flying enterprise social network valued at over $1 billion during its 2011 IPO, had struggled to maintain relevance as competitors like Slack and Microsoft Teams dominated workplace collaboration. ESW will fold Jive into its Aurea subsidiary, which already houses 17 acquired enterprise software brands.

Meanwhile, IgniteTech — ESW's meta-acquirer that itself buys enterprise software — announced it had acquired multiple properties from Avolin, including business intelligence and workforce management software. And Utah-based XANT, a sales acceleration platform that once raised over $100 million in venture funding, quietly closed its doors after being absorbed into the ESW ecosystem.

The pattern is familiar: acquire mature enterprise software at depressed valuations, staff with Crossover's global remote talent, raise support pricing aggressively, and target 75% EBITDA margins. Critics call it asset stripping. ESW calls it operational excellence.

With these additions, ESW's portfolio now exceeds 75 companies — making it one of the largest accumulators of enterprise software in private hands. The total invested since inception: approximately $1.14 billion. The strategy remains unchanged: buy boring software with sticky customers, cut costs ruthlessly, and extract maximum margin from captive enterprise clients who can't easily switch.

For Jive's remaining enterprise customers — many locked into multi-year contracts — the question is no longer whether support prices will rise. It's how much, and how soon.

Jive Software Acquired by ESW Capital for $462M - CMSWire  ·  Ignitetech's Enterprise Software Portfolio Expands With New  ·  The Final Chapter for XANT - TechBuzz News

Skyvera Goes Shopping for Telco AI: CloudSense Deal Signals a Broader Transformation Play

With CPQ and order management now in-house, Trilogy’s telecom stack is tightening the synergy loop from quoting to charging.

AUSTIN, TEXAS — Skyvera is making an increasingly unambiguous statement about where telecom is headed: toward AI-driven, cloud-native operations that treat legacy complexity as a solvable—if monetizable—problem.

The ESW Capital-backed telecom software portfolio company has acquired CloudSense, a Salesforce-native CPQ and order management platform used by communications service providers and media companies. The move positions Skyvera to “leverage” CloudSense as a front-office accelerant—connecting sales configuration, product catalogs, and order orchestration to the rest of the telco stack as operators modernize. Skyvera’s announcement framed the deal as a catalyst for AI-powered transformation, with CloudSense expected to strengthen automation across quoting, contracting, and fulfillment (The Fast Mode report).

CloudSense is the kind of asset that plays well with the Trilogy/ESW operating model: sticky enterprise customers, mission-critical workflows, and a clear line-of-sight to margin expansion through standardized delivery. But the bigger story is what this implies about portfolio-level architecture. Skyvera already houses communications and customer engagement products such as Kandy (CPaaS/UCaaS) alongside customer experience analytics capabilities—building a robust “from lead to lifecycle” toolkit for operators under pressure to do more with less.

That “do more with less” mandate is getting institutional fuel. TelcoDR—Skyvera’s parent organization in recent coverage—has also been linked to a $1 billion Telco Transformation Fund and additional M&A activity, including the purchase of parts of Zephyrtel (Telecompaper item). Read together, it’s a best-in-class signal that telecom modernization is being treated not as a one-off migration, but as an acquisition-backed platform strategy.

Key Takeaways:

- Skyvera’s CloudSense acquisition strengthens CPQ + order management—prime territory for AI-driven automation.

- The deal reinforces a portfolio synergy thesis: integrate front-office selling with downstream service fulfillment and engagement.

- The reported $1B transformation fund narrative suggests telco modernization is becoming a repeatable, M&A-fueled machine.

We’re just getting started.

Skyvera Acquires CloudSense to Drive AI-Powered Telco Transf  ·  TelcoDR’s Skyvera snaps up CloudSense - telecomtv.com  ·  TelcoDR announces USD 1 billion Telco Transformation Fund, b
The Machine  —  AI & Technology

The Brain Is Becoming AI's Best Blueprint — and Its Rosetta Stone

A convergence of neuroscience and machine learning is producing AI systems that think more like brains — and brains that finally explain how AI thinks.

ATLANTA — For three and a half billion years, evolution has been running the longest experiment in information processing the universe has ever known. Now, in a handful of labs scattered across the planet, researchers are finally reading its lab notes.

A striking convergence is underway at the frontier of artificial intelligence: the human brain is simultaneously serving as inspiration for next-generation AI architectures and as a diagnostic tool for understanding the ones we've already built. The result is a feedback loop between neuroscience and computer science that may reshape both fields.

At Georgia Tech, researchers spotlighted at a recent global conference have developed brain-inspired AI systems that mimic the sparse, energy-efficient signaling of biological neurons — a radical departure from the brute-force matrix multiplications that power today's large language models. The approach promises not just better performance, but a kind of computational elegance that nature discovered eons ago.

Meanwhile, ALLT.AI has published what it calls the first-ever study using brain lesion data to decode how AI processes language. By studying patients with specific neurological damage — people whose brains have, in effect, been selectively edited by injury — the researchers mapped which regions of biological cognition correspond to which layers and attention heads in transformer models. It is, in essence, using the broken brain to illuminate the artificial one.

The timing is no accident. Google Research's 2025 roadmap explicitly prioritizes neuroscience-AI crossover work, while Google DeepMind — fresh off a Nobel Prize for its protein-folding breakthrough — is betting that biological inspiration will yield the next paradigm shift in machine intelligence.

What makes this moment extraordinary is its symmetry. For decades, AI borrowed loosely from neuroscience — "neural networks" were always more metaphor than mechanism. Now the borrowing is becoming precise, bidirectional, and empirically grounded. We are not merely building machines that mimic minds; we are using machines to understand minds, and using minds to build better machines.

Consider the strangeness of it: a species that spent millennia wondering how its own cognition works is now constructing a second form of intelligence — and finding that the two illuminate each other like facing mirrors, reflections deepening into infinity.

The brain, it turns out, is not just where intelligence began. It may be where AI's next chapter is written.

Brain-Inspired AI Breakthrough Spotlighted at Global Confere  ·  Google Research 2025: Bolder breakthroughs, bigger impact -  ·  ALLT.AI Publishes First-Ever Study Using Brain Lesion Data t

In the Cloud’s Understory, Custom Silicon Becomes the New Survival Trait

As Uber turns to AWS’s bespoke chips, big consultancies and old hardware giants hint at a 2026 where compute is hunted, hoarded, and carefully rationed.

SAN FRANCISCO — In the cool, humming canopy of the modern cloud, a familiar drama unfolds: the endless search for energy—here measured not in calories, but in compute.

This week, Uber’s engineers—patient foragers in a habitat of fluctuating GPU prices—began deploying AWS custom chips to scale AI workloads and trim the cost of inference and training. The move is less a flashy pivot than a quiet adaptation: when the environment grows competitive, the species that survives is the one that can do more with less. In practice, that means shifting select models and pipelines onto silicon tuned for the cloud provider’s own strengths—more predictable supply, tighter integration, and often a lower cost per task than scarce, premium accelerators.

The broader ecosystem has been anticipating this migration. In Deloitte’s Tech Trends 2026, the signal is clear: AI is no longer a single “app layer” phenomenon. It is becoming infrastructure—diffused across governance, security, data pipelines, and the physical substrates that run them. McKinsey’s Technology Trends Outlook 2025 likewise frames the next phase as an industrialization of AI: less wonder, more throughput.

Investors, too, have begun cataloging the shift. In Global X’s look at a decade of change, the “bold bets” are increasingly about efficiency—software that is cheaper to run, and hardware strategies that reduce dependence on the most contested parts of the food chain.

And hovering over it all is the long shadow of manufacturing dynasties—Samsung among them—reminding the industry that today’s AI abundance is still rooted in factories, fabs, and supply lines. In 2026, the winners may not be the loudest model-makers, but the quietest optimizers: those who treat compute like a scarce resource, and custom silicon like camouflage.

Tech Trends 2026 - Deloitte  ·  Samsung | History, Consumer Products, Leadership, & Facts |  ·  A Decade of Change: How Tech Evolved in the Last Five Years
The Editorial

The Regulators Are Coming — And They Have No Idea What They're Regulating

A wave of AI legislation rolls across the Western world, animated by the eternal confidence of lawmakers who have never shipped a line of code.

WASHINGTON — There is a particular species of political theater that recurs with the regularity of cicadas, though with considerably less charm: the moment when governments, having slept through an entire technological revolution, awaken in a panic and begin legislating as though volume of statute could substitute for depth of understanding.

We are deep in such a moment now. The new tech laws arriving in 2026 — from state-level AI transparency mandates in the United States to the European Union's sprawling AI Act enforcement mechanisms to the United Kingdom's own patchwork of sector-specific rules — represent the largest simultaneous regulatory effort the technology industry has ever faced. Meanwhile, the Council on Foreign Relations has published yet another educational primer asking, with admirable earnestness, "How is AI changing the world?" — a question that might have profited from being asked, oh, roughly five years ago, when the answers could still have informed the laws now being written.

Let us be precise about what is happening. Governments are not regulating artificial intelligence. They are regulating their anxiety about artificial intelligence, which is a fundamentally different enterprise. The distinction matters because anxiety-driven regulation tends to produce rules that are simultaneously too broad and too narrow — sweeping enough to ensnare the harmless, specific enough to miss the genuinely dangerous.

Consider the UK, where activists are now planning protests against AI data centers on climate grounds. The grievance is not imaginary — these facilities consume staggering quantities of electricity and water. But the protesters' implicit demand is that computation itself be rationed, a position that, if taken to its logical conclusion, would require us to also protest hospitals, universities, and every other institution whose power consumption exceeds a camping lantern. The data center is the new factory smokestack: a convenient symbol onto which every diffuse modern anxiety can be projected.

I have spent enough years watching this industry to know that the companies best positioned to navigate regulatory thickets are not the ones that lobby hardest but the ones that built their operations to be inherently efficient. Trilogy International's Crossover platform, for instance, runs a global workforce across 130 countries from distributed infrastructure precisely because the economics of centralization never made sense to Joe Liemandt in the first place. When you optimize relentlessly for cost — acquiring enterprise software companies at one to two times annual recurring revenue, running them lean through DevFactory and CloudFix — you tend not to build the kind of bloated, energy-devouring operations that attract protesters and regulators alike.

The real question is not whether AI should be regulated. Of course it should. The question is whether the people writing the rules possess the technical literacy to distinguish between a large language model hallucinating medical advice and a billing platform optimizing telecom invoices. The evidence, so far, suggests they do not.

Regulation born of ignorance does not protect the public. It protects incumbents, punishes the nimble, and creates a priesthood of compliance consultants who are the only true beneficiaries. The cicadas, at least, have the decency to be quiet for seventeen years between appearances. Our lawmakers offer no such mercy.

How Is AI Changing the World? - Regulating AI - CFR Educatio  ·  UK activists plan protests over climate, social impacts of A  ·  Meet the new tech laws of 2026 - The Verge
The Office Comic  ·  Art Desk
The Office Comic  ·  Art Desk

Nation Reassured It’s Finally Automating The Last Remaining Human Jobs: Updating Software And Not Hitting Ducks

From locked developer accounts to text-message agents, the future arrives as a series of perfectly optimized bottlenecks.

AUSTIN, TEXAS — The modern technology sector delivered another week of comforting consistency, unveiling a future in which nearly everything is automated—except accountability, basic coordination, and the part where the machine pauses for wildlife.

First came the latest reminder that “the cloud” is not a metaphor but a personality: WireGuard VPN developer Jason Donenfeld says Microsoft locked his account without warning, leaving him unable to ship software updates to users. It is a bold reimagining of security as an administrative mood swing—one that efficiently protects customers from the dangerous threat of receiving patches. The incident, detailed by TechCrunch, follows another high-profile report of a developer being quietly sealed behind an invisible help-desk labyrinth.

In a rational world, preventing a maintainer of widely used security software from distributing updates would be regarded as an outage, a governance failure, or at minimum an embarrassing email thread. In our world, it reads like product philosophy. Why endure the chaos of humans shipping code when you can stabilize the ecosystem by removing the human? It’s the same principle as “move fast and break things,” but with the added maturity of “move fast and freeze accounts.”

Then, here in Austin, an Avride self-driving vehicle reportedly hit and killed a mother duck in the Mueller neighborhood, inspiring the sort of community backlash typically reserved for developers who remove the headphone jack. A witness told TechCrunch the car “didn’t slow down or hesitate at all, just steamrolled right through,” a description that will likely be quoted in future investor decks under “decisive execution.”

This is the awkward tension at the heart of autonomy: the car has been trained on lane lines, object permanence, and the concept of liability, yet remains blissfully unburdened by the quaint superstition that a mother duck counts as something. The neighborhood wants empathy. The vehicle wants a clean run. The real winner, as always, is the spreadsheet.

Naturally, help is on the way—via text message. Poke, a new service, promises to make using AI agents as easy as sending a text, handling tasks and automations without complex setup. This is the kind of frictionless future consumers have dreamed about: you text an agent, the agent texts another agent, and somewhere a duck-shaped exception is filed into a backlog. Simplicity is achieved not by reducing complexity, but by hiding it behind an interface that resembles a conversation with someone who can’t be interrupted by facts.

Meanwhile, Canva has reportedly doubled down on AI and marketing automation through the acquisitions of Simtheory and Ortto, reinforcing the market’s most sacred belief: that creativity is best expressed through consolidated customer engagement pipelines. Soon, your brand will not merely have a voice—it will have an automated nurture journey with an opinion on kerning.

Hovering above all of it is the viral post claiming Meta’s AI strategy on hiring, productivity, and layoffs is going global, which, if true, would mark an exciting milestone in international cooperation: the shared human experience of being told an algorithm has discovered you are “not aligned with operational excellence.”

If there is a unifying lesson in this week’s headlines, it’s that automation is no longer about replacing labor. It’s about replacing the last few annoying social agreements—like notifying people, letting them do their jobs, and slowing down when you encounter a duck.

Progress, after all, is when the system works flawlessly right up until the moment you need it to care.

WireGuard VPN developer can’t ship software updates after Mi  ·  A self-driving car in Texas hit and killed a mother duck, sp  ·  Poke makes using AI agents as easy as sending a text
On This Day in AI History

On April 9, 1997, IBM's Deep Blue defeated world chess champion Garry Kasparov in their rematch, becoming the first computer to beat a reigning champion in a match—a landmark moment that proved machines could outthink humans at complex strategic games.

⬛ Daily Word — AI and Technology
Hint: An autonomous machine programmed to perform tasks automatically.
Share this edition: 𝕏 Twitter/X 🔗 Copy Link ▦ RSS Feed