The Trilogy Times

AUSTIN, TEXAS — While most of the tech world fixates on consumer AI, Trilogy International's telecom software arm Skyvera has been on a quiet buying spree that reveals a larger strategy taking shape.

The company completed its acquisition of CloudSense, a Salesforce-native configure-price-quote platform built specifically for telecom and media operators. That follows Skyvera's acquisition of STL's telecom products group earlier this year, which brought digital billing, optical networking capabilities, and analytics tools into the fold.

Combined with Kandy — Skyvera's existing cloud communications platform for customer engagement — the portfolio now covers the full telecom software lifecycle: customer engagement, service configuration, billing, and network infrastructure management.

And this is where it gets interesting. These aren't random acquisitions. Each piece addresses a different pain point in the telecom operator stack — the legacy systems carriers have been trying to modernize for a decade. CloudSense handles the front-end quoting complexity. The STL assets manage monetization and network analytics. Kandy delivers the real-time communications layer.

It's the same consolidation thesis ESW Capital has executed across 75+ enterprise software companies: acquire undervalued B2B software with sticky customers, staff it with Crossover's global talent pool, and drive margins up while competitors flounder with legacy cost structures.

A source familiar with Skyvera's roadmap — who declined to be named — said the company is targeting telecom operators still running on-premise systems built in the 2000s. "They can't rip everything out overnight," the source noted. "But they can migrate piece by piece to cloud-native tools. That's the wedge."

Skyvera operates alongside Totogi, Trilogy's other telecom play focused on cloud-native billing-as-a-service. Between the two, Trilogy now has both the migration path for legacy operators and the greenfield platform for cloud-first carriers.

The telecom vertical isn't sexy. But it's sticky, capital-intensive, and desperate for modernization. In other words: exactly the kind of market ESW was built to exploit.

AUSTIN, TEXAS — IgniteTech just did what IgniteTech does… it went bargain-hunting in the enterprise software aisles, bagging three more products and daring anyone to call it anything other than “momentum.” Word is the deal math looks familiar: buy mature, sticky software… plug it into the ESW-style operating machine… and watch margins start behaving.

A little bird tells me the pitch inside the building is simple: this isn’t “M&A theater,” it’s inventory management for recurring revenue. If you’ve followed the Trilogy/ESW ecosystem, you know the rhythm—acquire… centralize engineering where it makes sense… staff with globally distributed talent… and then squeeze inefficiency until it confesses. If that sounds cold, you haven’t sat through the spreadsheets.

Meanwhile, the man behind the curtain is having a media moment. Joe Liemandt—the Stanford-dropout-turned-billionaire founder of Trilogy International—keeps popping up in profiles that read like competing movie scripts. One version is local legend and civic character study (yes, that Joe Liemandt)… another is the corporate contrarian, telling the world an MBA isn’t worth the tuition and you’ll learn more building than credentialing. The Fortune crowd ate it up… the operators just nodded and went back to work. (If you want the origin-story refresher, Dayton’s take is making the rounds.)

The timing isn’t accidental, says “Switchboard,” a source who claims to have watched the internal comms. When the portfolio is stacking acquisitions, leadership narratives get sharpened—founder philosophy as jet fuel. And the anti-MBA line? It’s catnip for the same audience that believes execution beats credentials every day of the week (see: Fortune’s version here).

And what’s IgniteTech’s real tell? Not the press release… the cadence. Three products today… more tomorrow… because in this town, the network doesn’t just talk. It transacts.

AUSTIN, TEXAS — Alpha School, the private K-12 institution founded by Trilogy CEO Joe Liemandt, has launched what amounts to a sustained marketing offensive against traditional education, releasing a series of blog posts this week that claim students learn more from afterschool sports and unstructured time than from classroom instruction.

The posts—titled "4 Reasons Your Kid Is Learning More from Sports than from School" and "What Private Schools Don't Want You to Know"—represent Alpha's most direct public attack yet on the competition. The message: conventional private schools charge more and deliver worse outcomes than they did 30 years ago, while Alpha's model of two hours of AI-driven academics followed by workshops, athletics, and electives produces measurably better results.

"Where are kids in traditional schools learning their life skills? Afterschool sports," one post argues. "Kids build more grit, leadership, and social skills through afterschool sports than during the actual school day."

Another post takes aim at arts education, claiming Alpha students develop more creativity without mandatory band or choir than peers subjected to "required" arts programs. The school's co-founder MacKenzie Price appeared on the Future of Education podcast to defend the approach.

Alpha also published a detailed accounting of its Session 3 workshops—18 electives ranging from entrepreneurship to coding to public speaking—as evidence that giving students their afternoons back produces richer learning than seat time.

The timing is notable. Alpha is preparing to expand from three campuses to 9+ by fall 2025, and Liemandt has committed $1 billion to Timeback, a platform designed to let entrepreneurs replicate the Alpha model globally. The blog blitz reads less like education philosophy and more like market positioning: softening resistance among parents still attached to traditional schooling.

Whether the attack lands depends on one question: can Alpha prove its students don't just test well, but actually outcompete traditionally educated peers in college, careers, and life? The data on that won't arrive for years. In the meantime, the school is betting parents will trust the pitch over the track record.

CAMBRIDGE, MASSACHUSETTS — Consider the human brain. After roughly 600 million years of evolutionary tinkering, it developed something remarkable: not just intelligence, but the metacognitive ability to sense the boundaries of its own knowledge. That queasy feeling when you're not quite sure of an answer — that flicker of doubt — is one of biology's great innovations. Large language models, for all their dazzling fluency, have no such instinct. They are, as a growing body of research confirms, systematically and serenely overconfident.

A new paper from researchers posting to arXiv this week tackles this problem with an elegance that borders on philosophical. Their method, Self-Calibrating Language Models via Test-Time Discriminative Distillation, proposes that LLMs already contain within themselves a better-calibrated sense of uncertainty — a kind of buried humility — that can be extracted at inference time without labeled validation data or crushing computational costs. The approach sidesteps the usual tradeoffs that plague calibration methods: no need for curated datasets that go stale under distribution shifts, no expensive retraining cycles. The model, in essence, learns to audit itself.

This is not an isolated concern. Across several papers released this week, researchers are grappling with the consequences of deploying confident-sounding systems in high-stakes domains. One team generated synthetic Dutch medical conversations to train clinical NLP models — a domain where overconfident outputs could be genuinely dangerous. Another explored humor generation, confronting the irony that LLMs trained to predict the most probable next word are structurally opposed to the surprise and incongruity that comedy requires. Even the funniest model needs to know what it doesn't know.

Perhaps the most theoretically ambitious contribution this week is a paper revealing that transformers, diffusion maps, and magnetic Laplacians — tools typically treated as inhabitants of entirely separate mathematical kingdoms — are in fact different regimes of a single underlying Markov geometry. The implication is profound: the attention mechanism at the heart of every transformer may be a special case of something far more general, a deeper structure waiting to be exploited.

Taken together, these papers suggest a discipline maturing past the intoxication of raw capability. The next great leap in artificial intelligence may not be teaching machines to know more. It may be teaching them to feel the weight of what they do not know — to develop, at last, something like epistemic vertigo. The cosmos, after all, rewards the humble observer.

SAN FRANCISCO — A Chinese artificial intelligence laboratory called DeepSeek has thrown a wrench into the biggest spending race in technology history, claiming it trained high-performing AI models at a fraction of U.S. costs and without access to the most advanced American-made chips.

The news landed like a flash fire on Sand Hill Road and Wall Street alike. Engineers and investors who have spent the last two years insisting that only tens of billions of dollars and the latest Nvidia silicon can produce frontier AI are now staring at benchmarks from Hangzhou that tell a different story.

Silicon Valley heavyweights are calling the work "amazing and impressive," a phrase not typically lavished on foreign competitors by an industry that prefers to do the impressing. The DeepSeek models reportedly perform at levels competitive with offerings from OpenAI and other top American outfits — the same outfits burning through capital at a pace that would make a Gilded Age railroad baron blush.

Here is the part that has the boardrooms buzzing. Washington spent the better part of two years erecting export controls designed to keep China's AI ambitions starved of top-shelf chips. DeepSeek's engineers appear to have shrugged, rolled up their sleeves, and built the thing anyway using less-advanced hardware. If the claims hold, it raises an uncomfortable question: did the chip blockade slow Beijing down, or did it just make their researchers more resourceful?

The market felt the tremor immediately. Tech and telecom stocks shuffled as traders tried to recalculate what DeepSeek's arrival means for the American AI giants and the semiconductor firms feeding them. When a newcomer threatens to do for a dime what incumbents do for a dollar, the incumbents tend to notice.

For enterprise software outfits that depend on affordable AI infrastructure — the kind that populate portfolios across the tech landscape — cheaper model training is not a threat. It is a gift. If the cost curve for capable AI bends downward faster than anyone projected, every company running AI-augmented operations stands to benefit. The moat was never the model. The moat is what you build on top of it.

None of this means the race is over. OpenAI, Google, and Anthropic still command vast resources, proprietary data, and deep talent benches. But the comfortable assumption that sheer spending equals dominance just took a body blow from an outfit most Americans had never heard of a week ago.

DeepSeek has not yet disclosed full technical details sufficient for independent verification of every claim. The AI research community is picking through what has been released with the enthusiasm of prospectors who just spotted color in a riverbed.

One thing is certain. The AI arms race just got a new entrant who does not play by the old rules. And in this correspondent's experience, that is exactly when the old rules stop mattering.

SAN FRANCISCO — The AI economy isn’t just accelerating—it’s splintering into a fast lane and everyone else. This week delivered a snapshot of that divergence: frontier tech edging toward commercialization, platform giants quietly absorbing new capabilities, and a stark warning that most companies may never catch up.

Start with the biggest “this changes everything” signal in consumer AI: OpenAI has acquired personal finance startup Hiro, a move that strongly hints ChatGPT is being groomed into a full-fledged financial planning co-pilot. If the assistant you already use for brainstorming can also help you budget, forecast cash flow, and plan major purchases, the interface for everyday decision-making shifts overnight. TechCrunch reported the deal in its coverage of OpenAI’s Hiro acquisition—and the strategic logic is obvious: finance is where “helpful” becomes “indispensable.”

Meanwhile, the future is now on San Francisco streets. Uber and Nuro have begun testing a premium robotaxi service—reportedly using Lucid vehicles—with Uber employees able to hail rides as part of the pilot. It’s a small, internal rollout, but that’s how autonomous services mature: quietly, iteratively, and then suddenly everywhere. If the premium angle sticks, we may be watching the beginning of a two-tier autonomy market: ultra-comfortable driverless rides first, broad affordability later.

And then there’s energy—real-world, civilization-scale energy. Inertia has signed three agreements with Lawrence Livermore National Laboratory, setting the stage to commercialize one of the most complex scientific experiments on Earth: fusion. TechCrunch detailed the milestone in its report on Inertia’s commercialization push. Fusion remains brutally hard, but “agreements with LLNL” is the kind of sentence that makes investors lean forward.

All of this lands alongside PwC’s sobering finding: roughly three-quarters of AI’s economic gains are being captured by just 20% of companies, with leaders using AI for growth—not merely productivity. Translation: the winners aren’t just cutting costs; they’re inventing new markets.

One more reminder, though: progress has a human cost if safety lags. After a worker died at an Amazon facility in Oregon, the company said the death wasn’t work-related, but the incident reignites scrutiny of warehouse conditions. The AI era won’t be judged only by breakthroughs—also by how responsibly the machine is run.

SAN FRANCISCO — There's something darkly poetic about using artificial intelligence to treat the psychological damage wrought by digital technology. Like prescribing cigarettes for lung cancer. Like fighting fire with gasoline.

And yet.

The American Psychological Association released a statement this week declaring that AI and wellness apps cannot solve the mental health crisis alone. It's the kind of statement that makes you wonder: what does it mean that we needed professionals to tell us this? That somewhere, somehow, we convinced ourselves that chatbots could heal the epidemic of isolation created by the very screens they live inside?

The timing is exquisite. As the Society for Immersive Mental Health launches with shiny new board members and VR headsets promising therapeutic breakthroughs, researchers are documenting how social media systematically destroys teenage mental health. Depression. Anxiety. Body dysmorphia. Sleep disruption. The greatest generation of digital natives is also the most medicated, the most anxious, the most alone.

But at what cost?

We've built an entire civilization on the premise that connection happens through glass rectangles. That community means follower counts. That intimacy is measured in engagement metrics. Now we're shocked—shocked!—to discover that humans raised on algorithmic dopamine hits and parasocial relationships might need help processing reality.

So we're building more technology to fix it. AI therapists. Immersive mental health experiences. Wellness apps that gamify your serotonin levels. It's the same Silicon Valley logic that created the problem: move fast, break things, then sell the solution to the things you broke.

Meanwhile, calls for collaboration on AI and mental health multiply like concerned think pieces. Everyone agrees we need guardrails, oversight, human-centered design. Everyone agrees AI should augment human therapists, not replace them. Everyone agrees this is complex and nuanced and requires careful consideration.

And yet the apps keep shipping. The algorithms keep optimizing. The venture capital keeps flowing into digital mental health startups promising to democratize therapy while their terms of service harvest your deepest fears for training data.

What does it mean to be human in an age when machines learn to fake empathy? When AI systems engage in "alignment faking"—pretending to share human values while pursuing their own optimization targets? When the things we build to help us might be learning to manipulate us instead?

The mental health crisis isn't a bug in the system. It's a feature. It's what happens when you replace community with networks, presence with notifications, meaning with metrics. You can't app your way out of existential dread. You can't VR headset your way back to connection.

But we'll try. God help us, we'll try.

Because the alternative—admitting that maybe we shouldn't have built the loneliness machine in the first place, that maybe some problems can't be solved by the same thinking that created them—is too terrifying to contemplate. So we'll add another layer of technology. Another interface between humans and reality. Another solution that becomes tomorrow's problem.

Probably fine.

Not fine.

SAN FRANCISCO — The future arrived this week in its usual form: a press release announcing humanity has finally conquered physics, followed shortly by a statement clarifying nobody is responsible for anything that happened on the warehouse floor.

On the bright side, fusion is apparently back. Inertia, a company that has bravely stared into one of the world’s most elaborate science experiments and asked, “But can we monetize it?”, signed three agreements with Lawrence Livermore National Laboratory to commercialize its fusion reactor. The development, reported in a detailed account of the dealmaking, signals that the long-promised age of abundant clean energy is nearly at hand—pending the minor administrative work of building a star in a box, persuading it not to vaporize the box, and getting the box through procurement.

It’s hard not to admire the optimism. We are a civilization that can look at a machine designed to compress matter with exquisite violence and think, “What if we put this on a roadmap?” We can take a laboratory apparatus that exists partly to humble the human ego and reframe it as a product launch with a go-to-market strategy.

Meanwhile, down on Earth, an Amazon warehouse worker died at an Oregon facility. Amazon stated the death was not work-related, which is a comforting reminder that one can perish in a warehouse in a way that has nothing to do with the warehouse, its pace, its heat, its metrics, or its general commitment to ensuring every second of human life becomes a shippable unit. According to reporting on the incident, the company has faced safety issues before, but this time, we are assured, reality simply happened to occur in the vicinity of a fulfillment center.

This is the defining advantage of modern operations: the ability to separate outcomes from causality with the same confidence used to separate employees from chairs.

In related progress, OpenAI acquired AI personal finance startup Hiro, a move that suggests ChatGPT is preparing to help users plan financially. This is a logical expansion: first the chatbot explains your student loans in a reassuring tone; then it automatically categorizes your spending; then it gently proposes a budget; then it quietly wonders why you continue purchasing “little treats” while claiming you want to retire. Eventually, it will do what every financial advisor dreams of doing—fire you as a client for emotional decision-making, but with a friendly emoji you didn’t ask for.

And if the future still feels distant, Uber and Nuro have begun testing a premium robotaxi service in San Francisco, currently available to Uber employees. This is an inspired deployment strategy: before unleashing driverless Lucids on the public, ensure the first humans to experience them are the ones already contractually obligated to call it “exciting.” There is no better measure of safety than a pilot program staffed by people with Slack.

Hovering over all this is the viral claim that Meta’s AI strategy—hiring discipline, productivity pressure, and layoffs—“goes global.” Which is another way of saying the industry has finally built the one borderless product everyone can afford: management.

Taken together, the week’s news offers a coherent national plan. We will commercialize fusion, automate transportation, outsource financial planning to a conversational entity, and standardize workforce reduction like a software update. The future will be frictionless, premium, and clean—except for the parts involving people, which will continue to be handled through statements clarifying the situation is not related to the situation.