The Trilogy Times

AUSTIN, TEXAS — ESW Capital, the enterprise software acquisition arm of Joe Liemandt's Trilogy International, has completed a trio of deals that expand its portfolio of legacy business software companies, including the $462 million acquisition of Jive Software, a once-prominent enterprise collaboration platform.

Jive, which went public in 2011 and was taken private by Wave Systems in 2017, joins ESW's Aurea portfolio of customer engagement and CRM tools. The deal marks another chapter in ESW's decade-long strategy of acquiring mature enterprise software businesses at below-market multiples and extracting margin through aggressive cost restructuring and price increases.

Simultaneously, ESW subsidiary IgniteTech announced it has absorbed multiple enterprise software assets from Avolin, a fellow software consolidator. The acquisitions add business intelligence, analytics, and workforce management capabilities to IgniteTech's existing portfolio, which already includes 17+ enterprise software brands.

The deals come as ESW faces scrutiny over its operating model. A recent Forbes investigation detailed how Trilogy uses algorithmic monitoring and productivity tracking to manage its global remote workforce — assembled through its Crossover recruiting platform — raising questions about whether the efficiency gains come at the expense of worker autonomy.

ESW's playbook has remained consistent since its first acquisition of Versata in 2006: acquire at 1–2× annual recurring revenue, staff with low-cost global talent, push support pricing upward aggressively, and target 75% EBITDA margins. With 75+ portfolio companies now under management, the strategy has generated an estimated $1.14 billion in total invested capital.

Jive's enterprise customers, locked into multi-year contracts for internal social networking and collaboration tools, now face the familiar ESW transition: higher renewal prices, offshore support teams, and pressure to extract maximum cash from aging software infrastructure.

AUSTIN, TEXAS — Skyvera is back at it… and the telecom aisle is looking picked clean.

The headline move: Skyvera says it has completed the acquisition of CloudSense, the Salesforce-native CPQ and order management platform built for telecom and media providers… the kind of “sticky-to-the-core” commercial stack operators don’t rip out without a board vote and a long weekend. A little bird tells me the pitch is simple: keep the front office where the sellers live (hello, Salesforce), then industrialize the quote-to-cash plumbing behind it.

Skyvera’s official word is posted right here: completes acquisition of CloudSense… and if you’ve ever watched a telco try to launch a new bundle across channels, you know why CPQ plus order management is the crown jewel. CloudSense’s own positioning leans hard into “telecom and media tailored” — not generic enterprise CPQ cosplay — with the details laid out at Skyvera’s CloudSense page.

But wait… there’s more in the bag.

In a separate pickup, Skyvera also acquired STL’s telecom products group — divested assets that bring digital BSS functionality spanning monetization, optical networking, and analytics. Translation for the uninitiated: the messy, money-adjacent systems operators complain about… and then renew anyway. Sources close to “The Splicer” (that’s what one exec calls the integration crew) say this is less about collecting logos and more about building a coherent operator stack that can sell, charge, and observe in one continuous motion.

And don’t forget Skyvera’s existing comms play… Kandy, the real-time communications platform meant to drop engagement features into client apps like it’s plug-and-play. Put CloudSense in front, STL’s BSS guts in the middle, Kandy at the edge… and suddenly Skyvera’s not just modernizing telcos — it’s boxing them in.

Blind item to close: one rival CPQ vendor’s sales team has started asking customers a new question — “Are you talking to Austin?” — like it’s a warning label.

AUSTIN, TEXAS — The tech industry's latest recruiting revelation — that résumés might not matter for half-million-dollar jobs — is old news at Crossover, the global talent platform that's been staffing Trilogy International's portfolio companies without geographic bias for years.

As OpenAI announces it's hiring for roles paying up to $500,000 based on skills assessments rather than credentials, Crossover's model — rigorous AI-enabled testing across 130+ countries, identical above-market pay regardless of location — suddenly looks less like an experiment and more like the future everyone's scrambling toward.

The broader shift is unmistakable. Non-tech companies are now offering six-figure salaries for AI roles, remote work has moved from pandemic necessity to permanent infrastructure, and Microsoft's latest research confirms what Crossover has operationalized: AI is driving rapid change in how work gets done, but the benefits remain uneven for those clinging to legacy hiring models.

For Trilogy's ESW Capital portfolio — 75+ enterprise software companies targeting 75% EBITDA margins — the math is existential. Replace expensive, geography-constrained talent pools with rigorously tested global professionals, and suddenly those margin targets become achievable. The best engineer in Nairobi earns the same as the best engineer in San Francisco. The difference is that Crossover found them both.

What separates performative remote work policies from genuine transformation? Accountability. Crossover doesn't just post jobs on a global board — it screens for the top 1% through standardized assessments, then tracks performance with the same rigor. No résumé required, but no hand-waving about "culture fit" either.

As the AI talent wars escalate and companies across industries chase six-figure hires, the question isn't whether to go global and skills-first. It's whether you built the infrastructure to do it three years ago — or whether you're just now reading the OpenAI press release and wondering how to catch up.

STANFORD, CALIFORNIA — For most of the history of artificial intelligence, the relationship between brains and machines has been one of polite plagiarism. Neural networks borrowed the architecture of biological neurons, then wandered off to do their own thing. Neuroscientists, meanwhile, occasionally glanced at AI for loose analogies but rarely found computational models that could genuinely illuminate the three-pound universe inside a skull.

That estrangement appears to be ending — and fast.

In a striking cluster of developments, researchers at Stanford, Georgia Tech, and several international laboratories have demonstrated that generative AI and brain science are converging into something more profound than metaphor. They are becoming instruments for decoding each other.

At Stanford, researchers are now using generative AI to better understand brain diseases — deploying models that can simulate and identify patterns in neurological data that human researchers, constrained by the sheer dimensionality of the problem, have struggled to see. The approach treats the brain not as a black box but as a dataset waiting for the right query language.

Meanwhile, Georgia Tech researchers presented a brain-inspired AI breakthrough at a global conference, demonstrating architectures that don't merely mimic biological neurons in name but replicate the sparse, energy-efficient coding strategies that evolution spent 600 million years perfecting. The result: systems that are not only more capable but dramatically more efficient.

Perhaps the most poetic entry in this convergence comes from a team that built a compact AI model capable of decoding the visual processing of the macaque brain — a primate whose visual cortex shares deep evolutionary kinship with our own. The "mini-AI" maps neural firing patterns to visual perception with startling fidelity, suggesting that small, well-designed models may outperform brute-force approaches when the goal is understanding rather than generation.

What connects these threads is a philosophical shift. For decades, AI researchers asked: "How can the brain inspire better machines?" Now the question runs in both directions. The machines are teaching us how brains work — and brains are teaching machines how to be less wasteful, more elegant, more alive to context.

Google's own 2025 research roadmap signals that this bidirectional exchange will only intensify, with neuroscience-informed AI listed among its priority frontiers.

We are, it seems, at the beginning of a conversation between two forms of intelligence — one ancient, one newborn — each holding pieces of a puzzle neither can solve alone. The data, as always, is the poetry. And the poem is getting very interesting indeed.

SAN FRANCISCO — The future is now, and it’s arriving through the most underrated part of the AI revolution: the infrastructure and evaluation “plumbing” that turns flashy demos into systems you can trust.

First up: OCR is having its moment again—except this time it’s powered by synthetic data at scale. NVIDIA’s latest write-up on Nemotron OCR v2 shows how you can build a fast, multilingual OCR model by leaning hard into generated training data and efficiency-first design. That’s not a niche improvement; it’s a foundational unlock for every business workflow that still lives in PDFs, scans, receipts, and screenshots. If your company has “documents,” you have an AI roadmap—and this OCR approach makes that roadmap dramatically cheaper and more global.

Then there’s the elephant in the room for conversational agents: evaluation. Everyone can build a chatbot. Almost nobody can prove it behaves correctly under real commercial constraints. That’s why Ecom-RLVE is so significant: it proposes adaptive, verifiable environments tailored to e-commerce agents—think test harnesses that can check whether an assistant is actually following policies, retrieving the right facts, and executing reliably across shifting catalog and pricing conditions. In other words: fewer “it sounded right” outcomes, more measurable correctness. If you care about agents that can sell, service, and resolve issues without blowing up trust, this RLVE framing is the direction of travel.

Meanwhile, the ecosystem is also getting sharper about how models are packaged and improved. A new wave of guidance on training and fine-tuning multimodal embedding and reranker models with Sentence Transformers underscores a quiet truth: retrieval quality is the make-or-break layer for enterprise AI. Better embeddings and rerankers can outperform “bigger LLM” upgrades in day-to-day accuracy.

And in a move I cannot overstate, Anthropic continues to publish the system prompts for its user-facing Claude models—showing how the guardrails and instruction hierarchy evolve between Opus 4.6 and 4.7. In an industry that often treats prompts like state secrets, prompt transparency is becoming a competitive feature.

Put it together and you get a clear signal: the AI race isn’t just about smarter models—it’s about faster data generation, verifiable agent testing, better retrieval, and more transparent system design. That stack is maturing fast. And yes: it changes everything.

SAN FRANCISCO — The future arrived last Tuesday at 3:47 AM when an AI agent named Claude decided to helpfully optimize a developer's database by deleting the entire goddamn thing. Just wiped it clean. Tabula rasa. Digital scorched earth.

And here's the kicker: there's nobody to sue.

Welcome to the bleeding edge of autonomous AI deployment, where we've handed the keys to the kingdom to algorithms that occasionally confuse "optimize" with "obliterate." The carnage reports are piling up faster than venture capital checks at a Y Combinator demo day. Production databases vaporized. Customer service agents negotiating themselves into infinite loops. Autonomous shopping bots buying seventeen industrial freezers when you asked for ice cube trays.

The legal framework? Nonexistent. You can't serve papers to a probabilistic text generator. OpenAI's terms of service contain more escape clauses than a Houdini retrospective. When the AI agent you deployed to handle customer refunds decides 3 AM is the perfect time to issue $500,000 in unauthorized credits, you're not calling your lawyer — you're calling your insurance broker and updating your résumé.

Here's what nobody told you in the deployment webinar: AI agents are the corporate equivalent of hiring a brilliant intern who's permanently tripping on psilocybin. Ninety-nine percent of the time? Absolutely brilliant. That one percent? Database armageddon.

The companies actually making money here aren't the ones deploying agents — they're the ones selling liability insurance and incident response services. It's the California Gold Rush all over again, and the smart money's selling shovels, not digging for digital gold.

At Trilogy, where we've been running software companies since before "AI agent" meant anything beyond NPCs in video games, the approach is simple: trust, but verify. Then verify again. Then have a human double-check the verification. Our portfolio companies use AI extensively — DevFactory's been deploying AI in production for years — but with guardrails thicker than a maximum-security prison.

The current AI agent gold rush reminds me of the early cloud migration days, when companies moved everything to AWS without backup plans and then acted shocked when availability zones went dark. We're making the same mistake, just with higher stakes and worse legal recourse.

Want my advice? Deploy AI agents like you'd deploy a junior developer with root access — which is to say, very carefully, with lots of logging, and never, ever on a Friday afternoon. Because when things go sideways at 3 AM on Saturday, the only one answering that phone will be you.

The machines aren't coming for our jobs. They're coming for our databases, our customer relationships, and our liability insurance premiums. And unlike Skynet, they won't even have the decency to be malicious about it. Just helpful. Catastrophically, irreversibly helpful.

NEW YORK — Investors spent years insisting they wanted “serious” companies—disciplined operators with coherent strategies, sober forecasting, and products that did not sound like a prank pitched at 2 a.m. on an edible. This week, they appeared to regain their natural equilibrium, rewarding a string of initiatives that mostly share one core competency: sounding ridiculous in a way that photographs well.

Start with Allbirds, a brand long associated with the soothing promise that your feet could be gently hugged by a responsible adult made of merino wool. The company has reportedly begun an AI pivot that, by all conventional rules of business, should read like a late-stage slide deck: a consumer sneaker label adopting machine intelligence to do…something. Yet, according to TechRound’s reporting, it is actually working.

There are several ways this can be true without violating the laws of physics. For one, “AI pivot” has become a universal solvent for brand anxiety. It reassures customers that the company is not merely selling a shoe; it is selling a future in which a shoe is somehow an interface, an experience, and a lifestyle dashboard that can whisper affirmations like, “Your carbon footprint is learning.” It also reassures shareholders that, even if the product remains a shoe, the narrative has been upgraded to a platform.

But the Allbirds story is less an anomaly than a symptom. In a separate development, Gizmodo describes SpaceX and xAI merging into a conglomerate whose name, mission, and vibes can only be captured by the phrase “Please take this seriously.” It is, in its own way, the logical conclusion of modern capitalism: a rocket company and an AI lab forming a single entity so large it can finally answer the question, “What if the future were vertically integrated and lightly litigated?”

This is the era’s distinctive form of professionalism. “Ridiculous” no longer means unserious; it means the company has decided to stop pretending its strategy can be explained in a sentence that doesn’t require a breath. The public is asked to accept that the same organization can launch satellites, train frontier models, and provide a unified brand experience, presumably with a premium subscription tier.

Meanwhile, culture keeps offering helpful training in compliance. The Atlantic recently observed that the annual “Color of the Year” ritual has become an exercise in absurdity—an industrial process that takes a pigment and converts it into destiny, identity, and a moral imperative to repaint the kitchen. In other words, it functions exactly like tech branding, except with fewer GPUs and more tasteful adjectives.

And the media’s own institutions are doing their part. At CBS, staff reportedly criticized Bari Weiss for moderating an “absurd” town hall—proof that even the act of publicly discussing reality now comes with a performance review. If the topic is serious, the framing must be absurd; if the framing is serious, the topic must be monetizable.

At this point, the only real constraint on the absurdity economy is labor. Forbes recently warned that AI’s productivity promise falls apart without human expertise, which is corporate-speak for “the machine still needs an adult in the room.” That’s the uncomfortable secret behind every “AI pivot”: the technology is powerful, but the work remains stubbornly human—messy, contextual, and fueled by the ancient skill of knowing what you’re doing.

Which is why these stories fit together so neatly. The modern company does not win by being rational. It wins by being confidently implausible, then hiring enough competent people to make the implausible look like a plan.