Contrary Research Rundown #136
Our new deep dive on the birth of the decentralized energy grid, plus new memos on Klarna, HeyGen, and more
Contrary is hosting NYC Tech Talk on June 9th, featuring eng leads from Clay, Graphite, Hebbia, and Moment. An evening built by engineers for engineers — each company will live demo their latest product features for leading builders in New York.
Please register to attend here!
Research Rundown
This week, we published an 11K+ word deep dive into the history of the electric grid and how its evolution has pushed the boundaries of a century-old system to the breaking point. In the piece, we unpack how the energy grid will have to adapt into a decentralized network of free-flowing energy — an internet for electricity.
America’s electric grid was once the marvel of the industrial age. A system that spans millions of individual generators, substations, transmission lines, and end users. A system that operates 24/7 with very little tolerance for downtime, all while incorporating century-old hydroelectric dams and analog transformers alongside real-time digital control systems, smart meters, and increasingly, AI-optimized distributed energy resources.
The electric grid of the 19th and 20th centuries was built around a “grow and build” model: more customers and bigger plants created a virtuous cycle of cheaper power and higher demand. By the 1920s, the US had thousands of power stations, and by 1930, nearly 70% of American homes had electricity. Post-WWII, America entered the golden age of the US power grid. Electricity use grew roughly 7% annually through the 1950s and 1960s, the market for electrical utilities became the largest industry in the country, and virtually 100% of US households had electric service by 1960.
But by the mid-1960s, cracks were starting to emerge. The 1965 Northeast blackout, which affected 30 million people, exposed a fragile, overly centralized system and marked the beginning of the end for unchallenged utility monopolies. In the 1970s and 1980s, a combination of environmental regulation, energy crises, and deregulation efforts reshaped the industry. Every effort in the latter part of the 20th century was, in effect, patching holes in what would eventually become a crumbling system.
Today, the way we create, store, transport, and use energy has changed dramatically. Thomas Edison and Nikola Tesla, the godfathers of the energy revolution in the 19th century, wouldn’t even recognize it. What was, originally, a system built for centralized, predictable power flows, is now being asked to juggle extreme peaks, bi-directional energy flows, and an explosion of new devices plugging in at the edge. Like trying to play 4K video on a VHS player. Together, these factors have created an escalating degree of fragility to the US electrical supply, which has already led to major outages.
If things continue as they are, there's worse to come. About 70% of transmission lines are over 25 years old, with many having exceeded their design life. A nationwide transformer shortage, fragile equipment, and a wave of utility retirements have made diagnosing and fixing grid problems harder, just as the grid’s complexity is exploding. This brittle system is increasingly giving way to meaningful energy demands. As of the end of 2023, 2.6K GW of proposed generation and storage projects were waiting for permission to connect to US grids, which is double the current installed capacity. But more than 70% of projects in queues never get built, throttled by long delays and interconnection backlogs.
These breaking points need to give way to the birth of a decentralized electric grid.
As pieces of the future grid come into place, stitching them together will require breaking old paradigms. It’s not just about adopting new technology, but about orchestrating an ecosystem of devices, software, and economic incentives in harmony. Think of how the internet evolved: open standards, smarter endpoints, and new business models transformed a patchwork network into a resilient web. The electric grid may be on the cusp of a similar transformation – an “Electricity Internet” where power flows as seamlessly as data, guided by intelligence at the edges.
You can read the full deep dive, “The Birth of The Decentralized Energy Grid,” here.
HeyGen provides an AI-powered video generation and translation platform that enables businesses and individuals to produce videos at lower cost, faster speeds, and for wider audiences. To learn more, read our full memo here and check out some open roles below:
AI Infrastructure Engineer - Los Angeles, San Francisco, Palo Alto, Toronto
Frontend Software Engineer - San Francisco, Palo Alto, Los Angeles, Toronto
Found is a fintech platform tailored for self-employed freelancers and SMBs to streamline financial solutions spanning banking, taxes, and accounting solutions. To learn more, read our full memo here and check out some open roles below:
Senior Machine Learning Engineer - San Francisco, Seattle, Portland, New York, or Remote
Senior Software Engineer (Full-Stack), Risk - San Francisco, Seattle, Portland, New York, or Remote
Poolside is building a generative AI foundation model, an API, and a coding assistant designed specifically for software engineers. To learn more, read our full memo here and check out some open roles below:
Member of Engineering (Applied Research Engineering) - Remote (US)
Member of Engineering (Back-end) - Remote (US)
Magical aims to disrupt traditional RPA by leveraging both no-code design and agentic AI. The platform allows users to create productivity shortcuts such as text expansion templates, auto-fill workflows, and AI-generated content in minutes, with no technical expertise or engineering resources needed. To learn more, read our full memo here and check out some open roles below:
AI Product Engineer - San Francisco or Toronto
AI Deployment Strategist - Toronto
Klarna is a global payments company and a dominant BNPL player, which has since expanded into banking, marketing, and commerce infrastructure. To learn more, read our full memo here and check out some open roles below:
Sox and ITGC Manager - London
Senior Analyst - Stockholm, Sweden
Check out some standout roles from this week.
Retool | San Francisco, CA - Software Engineer (AI Copilot), Software Engineer (Core Data), Software Engineer (Data Platform), Software Engineer (Developer Experience), Software Engineer (Observability)
Brex | Seattle, WA or San Francisco, CA - Engineering Manager (Frontend Platform), Senior Software Engineer (DEP), Software Engineer (Growth)
Cresta | Remote (US or Toronto) - Forward Deployed Engineer (AI Agent), Senior Full Stack Engineer (AI Agent), Senior Software Engineer (Backend - AI Agent)
OpenAI released HealthBench, a new benchmark for testing AI in realistic healthcare scenarios. Built with input from 262 physicians across 60 countries, it includes 5K multilingual, multi-turn conversations graded with physician-written rubrics. OpenAI says its latest models (like o3 and GPT‑4.1) outperform older systems and physician baselines in many areas.
Researchers at the Center for Genomic Regulation used AI to generate DNA sequences that precisely controlled gene expression in healthy mammalian cells—a first in synthetic biology. The AI-designed enhancers worked exactly as intended in mouse blood cells, enabling programmable gene activation across seven stages of development. This could unlock ultra-selective gene therapies with fewer side effects.
Figma has hired Morgan Stanley to lead its IPO, potentially one of the year’s largest. The $12.5 billion-valued design software company—whose $20 billion sale to Adobe was blocked by regulators—could go public later this year.
Legal AI startup Harvey is going multi-model and doubling down on client customization to defend its early lead in a now-crowded space. The $3 billion startup cited internal benchmarks showing that no single model dominates across all legal tasks—prompting a multi-model shift to better serve customers and power future legal AI agents. The company has $70 million in ARR, adoption by 8 of the top 10 US law firms, and a new Stripe veteran as chief business officer.
A new report from the National Security Commission on Emerging Biotechnology warns the US is falling behind China in critical biotech sectors. The Commission calls for urgent action to reshore manufacturing, reduce supply chain dependence, and ensure U.S. leadership in biotech innovation, positioning the field as central to national security and economic resilience.
The Boring Company hit a major milestone by continuously mining in a Zero-People-in-Tunnel (ZPIT) configuration for the first time. With no crew inside (except a videographer), the machine advanced while autonomously erecting 24,000-pound tunnel rings—an achievement likened to reusable rockets for its safety, speed, and cost efficiency.
Airbnb announced it has added Services and reimagined Experiences to its platform alongside a redesigned app. Guests can now book chefs, massages, workouts, and more across 260 cities, or join 650+ local experiences—from ramen classes to celebrity-hosted events—all within the Airbnb app, whether traveling or at home.
In response to Airbnb’s announcement, Pace Capital’s Chris Paik wrote an open letter to Airbnb CEO, Brian Chesky. In it, he argues that Airbnb’s Experiences product clashes with the company’s bottom-up DNA. He warns that top-down initiatives like curated activities misalign with Airbnb’s strengths—passive, scalable inventory driven by host behavior—and urges a shift toward organic, host-led offerings or a full pivot away.
Chime has filed for an IPO, aiming to capitalize on fintech’s market rebound. In 2024, the company reported $1.67 billion in revenue with a net loss of $25 million, significantly narrowing from a $203 million loss in 2023. With 8.6 million active members—67% of whom use Chime as their primary account—the company plans to expand into credit, investing, and insurance, while maintaining its identity as a technology firm, not a bank.
Anthropic is set to release new versions of Claude Sonnet and Opus that push “reasoning” AI further—allowing models to switch between thinking and tool use, self-correcting when stuck. The models aim to handle complex tasks with minimal input, signaling Anthropic’s continued bet on test-time compute despite mixed feedback on prior releases.
Perplexity is partnering with PayPal to launch agentic commerce, allowing users to shop, book travel, or buy tickets directly in chat starting this summer. The integration will use PayPal and Venmo for secure, in-flow checkout, aiming to make conversational commerce seamless across Perplexity’s AI-driven search platform.
Zepto announced its launch of Zepto Atom, a premium data insights subscription for consumer brands in India, offering granular, real-time analytics not available on other e-commerce platforms. Features include PIN code-level market share maps, minute-by-minute performance metrics, an in-house GPT assistant for strategic queries, and deep behavioral data to optimize pricing, marketing, and retention.
As AI competition intensifies, industry insiders say companies like Meta, Google, and OpenAI are prioritizing product launches over safety and research. Research groups like FAIR and Google Brain have been sidelined, safety testing timelines have been shortened, and experts warn that today’s most capable models are also more susceptible to misuse—with few safeguards in place to fix them.
Positive Sum's latest deep dive argues that software—not hardware—is the key bottleneck preventing robotics from reaching mass commercialization. Despite breakthroughs in humanoid pilots and AI models like π0 and Helix, the frontier remains constrained by limited real-world data. Investors are urged to focus on companies solving robotics data gaps, not just building robots.
Meta FAIR released new research artifacts advancing AI in molecular prediction, language processing, and neuroscience. Highlights include OMol25, the largest high-accuracy quantum chemistry dataset to date; UMA, a universal atom interaction model; Adjoint Sampling, a data-free generative training method; and a landmark brain-language study mapping how children develop language, revealing parallels with LLMs like Llama 3.1.
US officials have discovered rogue communication devices hidden in Chinese-made solar inverters and batteries, raising fears that firewalls could be bypassed to remotely disrupt the power grid. The findings, not previously disclosed, come amid growing scrutiny of China’s role in critical infrastructure, with some utilities already shifting away from Chinese suppliers.
A 9½-month-old boy with a deadly genetic disorder has become the first person to receive a fully personalized gene-editing treatment, marking a historic breakthrough in medicine. Developed to target his exact mutation, the one-time infusion appears to have saved his life and could pave the way for treating thousands of other rare genetic diseases.
Cohere, once seen as a top OpenAI rival, generated $70 million in annualized revenue earlier this year—falling 85% short of its 2024 forecast. Despite raising nearly $1 billion and a $5 billion valuation, the company struggled to commercialize its models and delayed a planned employee share sale. It’s now pivoting to enterprise apps like its new product North and doubling down on privacy-focused deployments.
In a recent TED Talk, Former Google CEO, Eric Schmidt, argued AI is still underhyped, warning of looming challenges like energy constraints, runaway agentic systems, and geopolitical escalation. He called for guardrails over halts, stressed the need for proof-of-personhood tech that preserves privacy, and painted a vision of radical productivity gains, personalized education, and AI-assisted healthcare—urging everyone to adopt fast or fall behind.
Takeout now accounts for 75% of restaurant traffic, with Gen Z and Millennials driving a cultural shift toward speed, convenience, and value, according to the National Restaurant Association’s 2025 report. Loyalty programs, AI ordering, and app-based deals are shaping habits, as younger consumers increasingly see takeout and drive-thru as essential to daily life.
At Contrary Research, we’ve built the best starting place to understand private tech companies. We can't do it alone, nor would we want to. We focus on bringing together a variety of different perspectives.
That's why applications are open for our Research Fellowship. In the past, we've worked with software engineers, product managers, investors, and more. If you're interested in researching and writing about tech companies, apply here!