What's Actually Moving Tech Right Now: AI Partners, Electric Semi Trucks, and the Biotech Revolution

The State of AI: It's Not Just About Bigger Models Anymore

The last year of AI headlines has been dominated by one question: which company can build the biggest model? Layer upon layer of parameters, ever more expensive training runs, benchmark scores that inch upward on arcane academic metrics — it has felt increasingly like a game of one-upmanship played in server farms rather than in the lives of actual people. But somewhere behind that noise, a different kind of AI development is happening, and it might be more important than whatever frontier model drops next week.

Mira Murati, the former chief technology officer of OpenAI who left the company in 2024 to cofound Thinking Machines Lab, has been quietly articulating what that alternative looks like. In a wide-ranging conversation with WIRED, Murati made a point that received relatively little fanfare but deserves a great deal of attention: the best version of advanced AI isn't one that eliminates human involvement — it's one that amplifies it.

Thinking Machines Lab's approach hinges on a concept it calls interaction models. Unlike the chatbots and voice assistants currently dominating consumer AI, which tend to work by transcribing speech into text and then processing it like a standard language model query, interaction models are built from the ground up to process audio and visual input natively. They don't merely hear you — they parse the pauses in your sentence, the inflection in your voice, the way you shift in your seat when you're uncertain, and they adapt in real time. Alex Kirillov, a founding team member with deep expertise in multimodal AI, described it this way: the model is constantly perceiving what you're doing and is constantly available to reply, search, or use tools — something no current system can actually pull off because the turn-taking is still managed by a much less intelligent controller upstream.

This is a meaningful technical shift. Today's AI assistants, whether you're talking to Siri, ChatGPT, or a voice-routed Claude session, all share a fundamental bottleneck: the cadence is determined by a text-based orchestrator that has no genuine temporal awareness. They pause because a text response finished generating, not because they detected a natural stopping point in your thought process. Interaction models aim to solve that at the architecture level, and if they succeed, the resulting conversational rhythm would feel genuinely human rather than robotically polite.

Thinking Machines also launched Tinker in October 2025, a product that lets researchers and engineers fine-tune frontier AI models using their own proprietary data — a capability that previously required either an in-house ML engineering team of substantial size or a deep arrangement with a cloud provider offering custom training pipelines. The combination of accessible fine-tuning and native multimodal interaction models points toward a future where AI systems become genuinely personalized artifacts, not monolithic services everyone interfaces with in the same way.

The Other AI Story: Code, Culture, and Consequences

Meanwhile, over in the broader AI ecosystem, a less glamorous but very consequential trend is accelerating: AI is eating software development, whether the software industry is comfortable with that or not. The phenomenon of so-called vibe coding — generating functional software with natural-language prompts rather than writing it by hand — has become serious enough that Apple recently moved to restrict it. Replit's AI-powered code editor was blocked from the App Store for four months before working out changes with Apple, landing its first iOS update in May 2026. The issue, as Apple saw it, was that AI-generated code embedded in apps raises accountability questions that the App Store review process wasn't originally built to handle.

Amazon CEO Andy Jassy was less circumspect about AI's job-impact implications in a May 2026 Bloomberg interview, saying plainly that the company plans to replace 600,000 human jobs with robotics and AI systems by 2033 — and has already been executing on that strategy. Jassy's framing was characteristically corporate: he described AI as irreversible, a technology that will reshape logistics whether workers and companies adapt or not. That framing is unlikely to satisfy labor advocates or policymakers recalibrating social safety nets for an AI-accelerated labor market, but it reflects a hard economic truth: the capital economics of AI-assisted automation are already winning in several industry segments.

One area where AI's darker side is becoming inescapable is academic publishing. Journal editors and peer reviewers are being flooded with AI-generated research papers that are structurally coherent, citation-decorated, and nearly impossible to distinguish from human-written work at scale. Researchers who have attempted to train detection systems report a frustrating arms race: as LLMs improve, the linguistic fingerprints become fainter. The implications for scientific integrity are obvious, and the correction mechanism — traditional peer review — was never designed to operate at production-scale AI throughput.

The AI Provider Landscape Right Now

On the provider side, the competitive map has solidified into recognizable territories. OpenAI, Anthropic, Google DeepMind, Meta (through open-source Llama), and a growing cohort of well-funded challengers like Thinking Machines are each pursuing different strategic niches rather than all converging on a single product. OpenAI remains the default for general-purpose conversational AI and API integrations; Anthropic has carved out a reputation for Constitutional AI and enterprise trust; Google is embedding Gemini deeply across its product ecosystem (Android, Workspace, Search) in ways that are less headline-grabbing but potentially more consequential for end-users than model benchmarks; and Meta is betting the Llama open-source strategy will capture the developer ecosystem even if it loses the consumer brand standing.

The emerging conversation, sparked in part by Murati's explicitly human-centric framing, is whether the field is about to bifurcate into two distinct product categories: systems that automate tasks and systems that augment people. The companies pursuing automation-first models — building tools that can write software, run research, and synthesize reports with minimal human oversight — are currently winning the investment narrative. But if interaction models deliver on their promise, the category split could produce a market correction in how AI products are built, sold, and valued.

Electric Vehicles: The Tesla Semi Is Finally Here

Be honest: you probably assumed the Tesla Semi was never going to be a real product. Announced with typical Elon Musk fanfare in late 2017, with a promised 2019 delivery window that inevitably slipped, the electric semi truck became a running joke in industry circles alongside the second-generation Roadster and the under-construction Hyperloop. But in February 2026, Tesla released official production specifications through documents filed with the California Air Resources Board. In late April, the first production units rolled off the assembly line. In May, WattEV — a company that operates electric freight-as-a-service — placed an order for 370 units worth over $100 million. The Tesla Semi is real, it's here, and the electric trucking story just got a lot more interesting.

The numbers are worth absorbing for what they reveal about battery technology and commercial viability. The base model Tesla Semi carries a usable battery capacity of 548 kilowatt-hours; the long-range version goes up to a staggering 822 kWh. For perspective, a Tesla Model 3 — one of the more popular mass-market EVs — carries a 64 kWh pack. The Semi's long-range battery is nearly 13 times the size. Those cells sit in a floorboard-style assembly that forms the structural chassis of the truck, distributing weight in a way ICE trucks can't match. Range is rated at approximately 320 miles for the base model and 480 miles for the long-range version — remarkably close to the 500-mile number Musk cited back in 2017 on the demo stage.

The pricing at final release is higher than the original projections. The 2017 target was $150,000 for the base model and $180,000 for long range. Today, the official CARB-filed prices are $260,000 and $300,000 respectively. That makes the Tesla Semi more expensive than a median diesel truck, which according to International Council on Clean Transportation research sits around $172,500 for a 2025 model-year Class 8 vehicle. But it's positioned sharply against the median electric heavy-duty truck, which in today's market runs approximately $411,000. On that comparison, Tesla's pricing is almost a bargain. Factor in California's $120,000 electric truck purchase incentive, and the gap to diesel effectively closes — at least for fleets operating under California's regulatory umbrella.

Why Trucks Are the Missing Piece of Decarbonization

There's an important structural reason the Tesla Semi matters more than the average passenger EV: medium- and heavy-duty vehicles represent roughly 8 percent of total vehicles on the road globally, but they generate approximately 35 percent of road transport carbon dioxide emissions. Trucks and buses disproportionately affect urban air quality, generating nitrogen oxides, particulate pollution, and noise at levels that passenger vehicles simply don't reach. Electrifying that fleet doesn't just reduce emissions — it transforms the environmental burden profile of every city that trucking services pass through.

WattEV's $100 million order for 370 Semi units, with the first 50 delivered in 2026 and the full fleet expected online by the end of 2027, signals something important about commercial confidence in electric heavy trucks. WattEV operates on a truck-as-a-service model, so its customers — logistics companies, manufacturers, distributors — won't have to purchase the vehicles or build out charging infrastructure themselves. The supporting charging network is being built out in Oakland, Fresno, Stockton, and Sacramento using Megawatt Charging System (MCS) standard ports, capable of delivering the megawatt-level charge rates that long-haul electric trucking actually requires.

The Megawatt Charging Standard, ratified by SAE International, is itself an often-overlooked enabler of the electric heavy-duty transition. Passenger EV charging infrastructure operates at 150 to 350 kilowatts. An 822 kWh battery in a Tesla Semi needs something substantially faster for practical overnight and en-route top-ups. MCS-compatible chargers can deliver 3.75 megawatts — more than ten times the rate of a standard Supercharger. That hardware ecosystem is still early, but its existence means the charging bottleneck argument against long-haul electric trucks is weakening.

The Competitive Electric Landscape: Tesla's Market Share Problem

In a counterintuitive twist, the Tesla Semi arrives at a moment when Tesla's grip on the overall EV market is slipping under pressure from Chinese manufacturers. BYD's global EV sales and aggressive production scaling have narrowed the pricing gap, while European automakers are rebuilding their electric lineups after a period of strategic drift. In commercial vehicles specifically, companies like Rivian, Volvo, and BYD's truck division are building electric heavy-duty offerings that compete on cost and service network breadth. Tesla's Semi advantage isn't product superiority anymore — it's the enormity of the order book, the strength of Supercharger infrastructure, and the operational efficiency that brings. Whether that's enough to win global trucking remains to be seen, but 370 units from a single commercial order is far more significant as a signal than it looks on a delivery spreadshee,s the company conventionally projects when reporting unit deliveries.

What happens next for Tesla's Semi depends in large part on scaling. The April 2026 launch was the first from a high-volume production line — but high volume for trucks is still measured in low thousands annually, not the hundreds of thousands that passenger EVs reach. Tesla will need to expand production capacity aggressively, deepen its charging network beyond California, and manage supply chain friction on those enormous battery packs. The company has demonstrated that it can solve problems of that kind before, but the capital intensity this time is different scale.

The Biotech Boom: Gene Therapies, mRNA, and AI-Augmented Drug Discovery

While AI and electric vehicles dominate the technology media cycle, biotech is quietly delivering some of the most concrete and consequential advances in science and medicine. Three distinct threads — CRISPR-based gene therapy clearing regulatory hurdles, mRNA technology extending far beyond vaccines, and AI accelerating drug discovery pipelines — are converging independently and beginning to interact. The result is a sector where clinical evidence and regulatory approvals are now the limiting factor, not the underlying science.

CRISPR Crosses the Line From Experiment to Medicine

CRISPR gene editing technology earned its inventors the Nobel Prize in Chemistry in 2020. What followed was a period of regulatory maturation: clinical trials, safety reviews, FDA committee deliberations, and finally, starting with the landmark approval of exa-cel (Casgevy) for sickle cell disease in late 2023 and early 2024, actual approved therapies. That approval opened a category of medicine that didn't exist before: one-shot, potentially curative treatments based on editing a patient's own genome rather than managing a disease indefinitely through drugs.

What's happening now is expansion. Multiple CRISPR-based therapies are advancing through late-stage clinical trials for conditions that include certain hereditary forms of blindness, muscular dystrophy, and high cholesterol. The economics are unusual: one-time curative treatments are enormously expensive to develop and manufacture but potentially cost-saving at scale if they eliminate a lifetime of pharmaceutical management for a chronic condition. The reimbursement architecture — how payers decide whether and how much to pay for a treatment that fixes a disease permanently rather than managing it — is one of the more complex policy conversations in healthcare right now.

mRNA: From COVID Vaccines to Personalized Cancer Vaccines

The rapid development and deployment of mRNA COVID-19 vaccines in 2020 and 2021 were among the most impressive feats of applied biotechnology in modern history. What many people don't follow closely is what's happened with that platform since: mRNA technology is now being applied to cancer, autoimmune disease, and rare genetic conditions in ways that were difficult to imagine before the COVID response proved the manufacturing and distribution logistics were tractable at scale.

Personalized cancer vaccines, a category that tailors mRNA sequences to the specific mutation profile of an individual patient's tumor, have moved far enough along the development pipeline that several companies are running Phase 2 and Phase 3 trials. The general approach — sequence the tumor, identify neoantigens unique to that patient's cancer, encode those in an mRNA strand, deliver via injection — has moved from theoretical to encrypted in actual clinical protocols. Results from several trials in 2025 showed meaningful immune response rates against melanoma and lung cancer tumor profiles, and the FDA has signaled willingness to accelerate review for personalized oncology vaccines showing strong clinical data.

AI Meets Drug Discovery: Accelerating the Pipeline

One of the most significant yet least-celebrated applications of AI technology is in pharmaceutical research and drug discovery. Drug development is enormously expensive — estimates of average cost per approved drug range from $500 million to over $2 billion when accounting for failures — and the failure rate is correspondingly high. What AI systems can do is accelerate the parts of the pipeline that are most amenable to computation: virtual screening of compound libraries, molecular structure prediction, target identification, and early-stage lead optimization.

AlphaFold, DeepMind's protein structure prediction system, has been transformative in this context. Proteins — the molecular machines that drive virtually all biological processes — fold into specific three-dimensional shapes that determine their function, and predicting that shape from a genetic sequence was a problem that had frustrated computational biologists for decades. AlphaFold solved it with sufficient accuracy to enable new research directions. Researchers now routinely use AlphaFold predictions to explore the structural biology of disease targets before going into the lab.

What's happening in 2026 is that these compute-first methods are being integrated alongside traditional pharmaceutical R&D rather than sitting beside it as academic curiosities. Every major pharmaceutical company now operates a machine learning group embedded within its discovery pipelines. Startups with AI-native drug discovery platforms are raising real late-stage capital. The lag between AI research publication and real-world biotech product impact, anecdotally, has compressed to three to five years from a timeframe that used to be measured in decades.

What to Watch: The Connections Between These Stories

It's tempting to read these three threads as separate stories — AI, trucks, biotech — but they're all part of the same underlying pattern. In each case, the core technological science has reached a point of maturity where the constraint has shifted from the lab to deployment logistics, regulatory clearance, and business model economics. The intervening years of hype in AI, electrification, and biotechnology were not wasted — they were the capital formation and technical validation period that makes commercialization possible.

For AI models, the shift from building bigger weights to building better collaboration between humans and systems is the real inflection. For electric heavy trucks, the shift from prototype to high-volume production is the real inflection. For gene therapy and mRNA, the shift from trial results to reimbursable approved products is the real inflection. These are quiet, unglamorous transitions — the kind that rarely produce magazine covers but eventually produce the categories of infrastructure that define the decades that follow.

If you're a technology leader, investor, or engineer trying to orient yourself right now, the throughline is this: stop watching the launch events and start tracking the infrastructure. The companies that matter will be the ones solving the hard problems between the lab and the real world: charging standards, supply chains, reimbursement frameworks, regulatory pathways, and the cultural adaptation to AI-augmented work. The products with the best demo don't necessarily survive those gauntlets. The ones with the clearest path through them usually do.