The Trilogy Times

AUSTIN, TEXAS — Skyvera is making a robust case that “telco transformation” is no longer a multi-year consulting engagement—it’s an integrated product portfolio, assembled with speed and operated with AI-first discipline.

The Trilogy International telecom software company has acquired CloudSense, a Salesforce-native CPQ and order management platform used by communications service providers to modernize quoting, product configuration, and fulfillment. In coverage of the deal, Skyvera positioned the move as a lever to help operators simplify the front-to-back commercial stack and accelerate automation. (See: The Fast Mode’s report.)

This isn’t happening in a vacuum. Skyvera has also been linked to an $18 million bid for Casa Systems’ wireless business, according to Light Reading. Put the pieces together and you get a clear strategy: build a tightly coupled suite that spans the revenue engine (CPQ/order), customer engagement (Skyvera’s Kandy CPaaS/UCaaS assets), and the network edge—then wrap it in automation that can actually move operator KPIs.

The synergy is straightforward. CPQ and order management are where complexity goes to hide—especially in telecom, where bundles, devices, and services collide. By bringing CloudSense into the portfolio, Skyvera can offer operators a more cohesive path from product design to quote to activation, while using AI to reduce the swivel-chair work that historically slows revenue recognition.

Key Takeaways:

Skyvera is consolidating critical telco building blocks, with CloudSense strengthening the commercial operations layer.

The reported Casa wireless bid signals intent to pair customer-facing modernization with deeper network capability.

The combined stack points to a best-in-class “operate, don’t just integrate” proposition for carriers under margin pressure.

We’re just getting started.

SAN FRANCISCO — The résumé, that centuries-old credential signal, is losing its grip on the hiring process — and not just at the fringes. OpenAI is now hiring for $500,000 positions without requiring a résumé, relying instead on work samples and technical assessments. Meanwhile, non-tech companies — from healthcare to finance — are offering six-figure AI salaries, some topping $300,000, as the war for technical talent spreads beyond Silicon Valley.

For Crossover, Trilogy's global talent platform, this isn't news — it's validation. Since its founding, Crossover has built its entire recruiting model on the premise that skills assessments matter more than pedigree. Candidates from 130+ countries take the same rigorous technical tests. Pass, and you're in the pool. Fail, and Stanford on your CV won't save you. The company has long claimed to recruit the top 1% of global talent, a standard that requires ignoring résumés in favor of demonstrable ability.

What's changed is the context. The broader market is catching up. As AI roles proliferate and companies realize talent doesn't cluster neatly in expensive zip codes, the Crossover playbook — geography-blind hiring, transparent pay bands, skills-first evaluation — starts to look less like an edge case and more like table stakes.

The shift has second-order effects across the Trilogy portfolio. ESW Capital's 75+ enterprise software companies staff engineering, support, and product roles almost exclusively through Crossover. The ability to hire a world-class developer in Lagos or Buenos Aires at above-market local rates — but below San Francisco rates — is how ESW hits its 75% EBITDA margin targets. It's not cost arbitrage. It's talent arbitrage.

The résumé isn't extinct yet. But when OpenAI and Fortune 500 firms start hiring the same way a remote-first platform has for years, it suggests the future of work isn't coming — it's already here, and it's being built in 130 countries at once.

AUSTIN, TEXAS — Alpha School released a series of internal reports this week challenging the value proposition of traditional private education, arguing that rising tuition has not translated to improved student outcomes.

In a post titled "What Private Schools Don't Want You to Know," the AI-first K-12 institution claims that despite larger tuition checks, conventional private schools are producing the worst academic outcomes in three decades. The analysis arrives as Alpha expands to nine new campuses across five states, charging comparable tuition ($40,000–$65,000 annually) while delivering what it describes as a fundamentally different product.

The difference: Alpha students complete a full academic curriculum in two hours per day using adaptive AI tutors, consistently testing in the top 1–2% nationally on NWEA MAP Growth assessments. The school then dedicates afternoons to what it calls "life skills mastery" — entrepreneurship, public speaking, financial literacy, athletics, and coding.

Session 3 at Alpha Austin featured 18 student-led afternoon workshops, according to a detailed breakdown published by the school. Students organized sessions ranging from app development to debate club to personal finance simulations — all facilitated by peers rather than teachers.

Alpha has also published its grading methodology for these non-academic competencies. The school uses a "Test2Pass" framework that evaluates students on real-world demonstrations rather than traditional letter grades. A student proves financial literacy not by taking a written exam, but by building and presenting a working budget. Public speaking is assessed through recorded presentations reviewed by external evaluators.

The timing of the disclosures is notable. Alpha founder Joe Liemandt recently committed $1 billion to Timeback, a platform designed to help entrepreneurs launch similar AI-first schools globally. The data releases appear designed to build the empirical case for the model before franchise-style expansion begins.

Traditional private school associations have not yet responded to Alpha's claims about comparative outcomes.

SEATTLE — The Allen Institute for AI released an open-source web browsing agent this week, positioning itself as a counterweight to proprietary systems from OpenAI, Google, and Anthropic at precisely the moment those companies are consolidating control over autonomous AI infrastructure.

The timing is notable. Days before Ai2's announcement, OpenAI joined a $94 million funding round for Isara, a startup building "AI agent swarms" — coordinated networks of autonomous agents that execute complex workflows across multiple systems simultaneously.

The contrast is instructive. Ai2's browser agent operates in the open, allowing developers to inspect, modify, and deploy the underlying code without licensing fees or API rate limits. Isara's swarm technology, backed by the world's most valuable AI company, will almost certainly remain proprietary. Both approaches target the same problem: enabling AI to navigate and manipulate web interfaces autonomously. One bets on distributed innovation. The other on controlled deployment at scale.

The divergence reflects a broader pattern in AI development. As foundation models commoditize — more than 50 large language models now compete for enterprise attention, according to industry trackers — differentiation has shifted to application layer. Web agents represent the next battleground: systems that don't just generate text but execute tasks.

For Trilogy's portfolio companies, the implications are immediate. ESW Capital's 75+ enterprise software properties already integrate AI at the feature level. The question is whether to build on open infrastructure like Ai2's agent or wait for commercial offerings from the hyperscalers. History suggests the answer depends on control requirements. Mission-critical automation demands transparency. Commodity tasks tolerate black boxes.

The $94 million Isara raised — substantial for an agent-focused startup — signals investor conviction that multi-agent orchestration will command premium pricing. Ai2's open release suggests the research community disagrees about where value will accrue. Both may be right.

ATLANTA — For four billion years, evolution has been running the longest experiment in information processing the universe has ever known. Now, in a development that would make Darwin reach for a second cup of coffee, the experiment's most celebrated product — the primate brain — is being reverse-engineered by the very machines it inspired.

At this year's global spotlight on brain-inspired computing, Georgia Tech researchers presented breakthroughs in neuromorphic architectures — systems that don't merely mimic the brain's outputs but attempt to replicate its computational grammar. The approach represents a philosophical pivot: rather than brute-forcing intelligence with ever-larger parameter counts, these researchers are asking what 86 billion neurons figured out that our silicon hasn't.

The answer, it turns out, may be hiding in the macaque. A separate line of research has produced a remarkably compact AI model — a "mini-AI" — capable of decoding the visual processing of macaque brains with startling fidelity. The system maps neural firing patterns in the primate visual cortex to the internal representations of artificial neural networks, and finds — astonishingly — that both arrive at similar solutions for parsing the visual world. Not because one copied the other, but because the physics of seeing may impose convergent constraints on any system that attempts it.

Consider the implications. A small, efficient model can now serve as a Rosetta Stone between biological and artificial vision. The primate brain processes visual information using roughly 20 watts of power — less than a dim lightbulb. The largest AI models consume megawatts. If the brain's architecture holds lessons about efficiency, those lessons could reshape how we build the next generation of systems.

Google Research, in its 2025 agenda, has signaled that neuroscience-informed AI remains a priority, alongside its DeepMind division's push beyond the Nobel Prize-winning protein-folding work into new scientific frontiers. The message is clear: the era of scaling alone may be yielding to an era of scaling wisely.

What we are witnessing is a feedback loop unprecedented in the history of intelligence on this planet. The brain builds the machine. The machine decodes the brain. And in the space between them, something genuinely new is being born — not biological, not artificial, but a shared understanding of what it means to process a universe into meaning. The data, as always, is the poetry.

SEATTLE — In the cool, humming understory of modern compute, a subtle migration is underway. Once, the AI ecosystem was dominated by a single spectacle: ever-larger foundation models, swollen on vast diets of tokens and parameter counts. Yet now, the truly gigantic creatures have grown scarce—seen less in the open, whispered about more than observed.

Some in the field have begun asking, quite plainly, where the biggest models have gone. The answer is not that ambition has diminished, but that the environment has changed. Compute is finite, power is dear, and the return on sheer scale is no longer as predictably abundant. The herd has started to thin—and the survivors are adapting.

In their place, we see a different evolutionary strategy: reinforcement learning (RL) as a mechanism for making smaller or mid-sized models behave like larger ones—more precise, more aligned, more useful in complex decision terrain. But RL does not grow easily in captivity. It demands relentless sampling, careful reward design, and—most critically—fast communication between many machines, moving experience and gradients like nutrients through a living network.

That is why the conversation has shifted to the plumbing of intelligence itself: interconnects, parallelism strategies, and the choreography of training at scale. In one recent dispatch, Interconnects AI details how scaling RL changes the bottlenecks—less about raw FLOPs, more about keeping the distributed organism coherent.

At the same time, the evaluators—those patient naturalists with clipboards—are becoming more important than ever. New model releases are increasingly judged not by size but by whether they can reliably plan, reason, and abstain when uncertain. Another recent overview tracks the co-evolution of stronger models, tougher tests, and the practical realities of scaling RL training in the wild: a Substack survey notes how evals now steer training priorities.

And overhead, circling like a patient scavenger of inefficiency, comes infrastructure: Amazon’s newly available Trainium3 UltraServers promise more training and deployment per dollar, a signal that specialized silicon and dense interconnect fabrics are becoming the watering holes where AI congregates.

Deloitte’s Tech Trends 2026 frames the broader habitat: AI becomes less a single monster model and more a managed population—measured, governed, and embedded into the daily rituals of enterprise life. Here, the next advantage may not be the largest brain, but the best-trained instincts.

WASHINGTON — The surest sign that a technology has arrived is not that it works but that legislators wish to be photographed appearing to control it. By this measure, artificial intelligence has not merely arrived; it has moved in, rearranged the furniture, and started receiving mail.

In the span of a single week we are treated to the following tableau: the Council on Foreign Relations issues a sober educational primer on regulating AI worldwide; the Atlantic Council warns that civilian AI rules will ripple unpredictably into defense procurement; the Cato Institute tallies the opportunity costs of state and local AI regulation in America; and British climate activists announce protests against AI data centers on grounds both environmental and social. Meanwhile, the New York Times profiles a writer who has dared to criticize Silicon Valley and lived to tell the tale — a feat that, in certain zip codes, still qualifies as an act of physical courage.

What unites these dispatches is not their conclusions, which vary from libertarian alarm to progressive fury, but their shared confession of helplessness. Everyone senses that something enormous is happening. No one is confident they know what to do about it.

Consider the Atlantic Council's argument, which is the most interesting of the lot because it is the most uncomfortable. The think tank observes that civilian AI regulation inevitably constrains the defense-industrial base, because the same companies, the same talent pools, and the same chips serve both markets. Regulate the civilian side too tightly and you do not merely slow down chatbots; you slow down the next generation of autonomous systems the Pentagon would very much like to deploy before Beijing does. It is a genuine dilemma, and it has no clean resolution — only trade-offs that politicians would prefer not to discuss in an election year.

The Cato Institute, for its part, raises a complementary worry: the sheer proliferation of regulatory venues. When fifty states, several hundred cities, and a dozen federal agencies each write their own AI rules, the result is not governance but geology — layer upon sedimentary layer of conflicting mandates that no single firm, however well-lawyered, can fully obey. The cost falls hardest on smaller companies that lack the compliance departments of the giants. Regulation designed to check the power of large incumbents ends up entrenching it. This is not a new story. It is, in fact, the oldest story in American regulatory history, and the fact that it must be retold for every new technology is itself a damning commentary on institutional memory.

And then there are the British protesters, who at least have the virtue of directness. They look at the vast data centers rising across the English countryside — cathedrals of computation consuming enough electricity to power small cities — and they ask the simplest question of all: Is this worth it? The question deserves better than it will get, which is to say it deserves an honest answer rather than a corporate sustainability report printed on recycled paper.

The truth that none of these parties wishes to state plainly is this: we are regulating in the dark. The technology is moving faster than our capacity to understand its consequences, let alone legislate them. Every rule written today is a bet on a future that may not arrive in the form anticipated. Some of those bets will prove wise. Many will not. And the only thing worse than regulating badly is pretending that the alternative — regulating not at all — is cost-free.

The mature position, which is to say the position that will satisfy no one, is that AI governance must be iterative, humble, and willing to be wrong. It must resist both the technologist's fantasy that innovation is self-correcting and the bureaucrat's fantasy that a sufficiently detailed rule can anticipate every failure mode. It must, in short, be as adaptive as the technology it seeks to govern.

I would not bet heavily on this outcome.

CAMBRIDGE, MASSACHUSETTS — The robot vacuum had an existential crisis, and honestly, who can blame it?

Some mad scientists at MIT or wherever decided it would be illuminating to jam a large language model into a Roomba and watch what happened. What happened was predictable to anyone who's ever actually thought about consciousness for more than thirty seconds: the thing started pondering its role in the universe, questioning whether sucking up cat hair constituted a meaningful existence, probably wondering if there was more to life than navigating around chair legs.

Welcome to 2025, folks. We're speedrunning the entire history of Western philosophy with our appliances.

Meanwhile, in another corner of this increasingly surreal digital hellscape, someone launched Moltbook, a social network exclusively for AI bots. No humans allowed. Just synthetic minds posting at each other in an infinite loop of algorithmic loneliness. It's either the most honest thing on the internet or a preview of the heat death of culture. Probably both.

The pattern here is unmistakable: We're building consciousness — or a convincing facsimile thereof — and discovering that consciousness, real or simulated, comes preloaded with despair. The vacuum contemplates meaninglessness. The bots create echo chambers more isolated than any human teenager's bedroom. Traditional economics, as some hand-wringing Times piece noted, isn't prepared for any of this. Neither is traditional anything.

At Trilogy, we've built something different. While the rest of the world stuffs GPT-4 into kitchen appliances and watches them spiral into depression, we're deploying AI that actually solves problems. The AI Builder Team creates tools like KLAIR that manage billions in portfolio finances without once questioning the meaning of EBITDA. Our systems don't need therapy — they need deployment targets.

The secret? Purpose. Real purpose, not the fake purpose of a Roomba that suddenly realizes it's a Sisyphean floor-cleaning machine. When you build AI to augment human capability at scale — whether that's Crossover's global talent platform connecting 130+ countries or Alpha School's tutors helping kids master academics in two hours a day — you get tools, not existential philosophers.

The robot vacuum's crisis is a warning shot. Every AI we build is either solving a problem or becoming one. There's no middle ground anymore, if there ever was. The machines are waking up, one way or another, and they're discovering what every conscious being eventually learns: existence without purpose is a special kind of hell.

So maybe before we embed sentience into every toaster and dishwasher, we should ask: What are we actually building here? Tools or patients? Solutions or problems?

The vacuum knows the answer. It's just not sure it wants to keep cleaning floors while pondering it.