The Frontier

Diogo Almeida argues coding agents like Cursor or Claude Code merely speed up the production of traditional code without making software smarter. Jev provides a new programmatic primitive that integrates natural language intent directly into program state decisions.

Diogo Almeida describes Jev as a highly advanced classifier designed to function like a database or standard library within software systems. It accepts natural language inputs to guide state machine decisions, avoiding floats where AI predictably struggles.

TypeSafe prioritizes intelligence per dollar over speed as its primary design metric. Diogo Almeida explains this focus is necessary to ensure AI can be integrated deeply into the internal state of critical software systems rather than just human-facing chat layers.

Diogo Almeida entered the AI field by winning a Kaggle competition using system automation and nested loops rather than advanced mathematics. This victory led SVM co-inventor Isabel Guyon to introduce him to the broader machine learning research community.

At OpenAI in late 2021, Diogo Almeida realized RLHF generalization was real after testing the model with the query "why is it important to eat socks before meditating." The model generated plausible, human-like answers for the unseen query.

OpenAI has attempted to automate customer service since 2020 without achieving robust reliability. Diogo Almeida attributes this failure to the industry optimizing models for human judges and impressive demos rather than background task execution.

Diogo Almeida rejects the idea that AI is on a path toward recursive self-improvement. However, he believes automating economically valuable work is highly achievable because the majority of real-world work consists of simple, rote instructions.

The Saas-pocalypse theory failed because complex software architecture is extremely difficult to replicate beneath the surface. Diogo Almeida predicts an inverse Saas-pocalypse where established SaaS companies become the biggest AI winners by deeply automating existing customer workflows.

Martin Casado notes a Google study revealed the average pull request at a large tech company is only ten lines long. This suggests that accelerating syntax generation via AI agents targets a minor bottleneck in software development.

Diogo Almeida anticipates a new era of probabilistic programming where developers trade off intelligence, cost, and speed. Systems engineers will use lightweight, approximate guesses to dynamically route data within complex infrastructure.

Harvey's valuation surged from $3 billion in February 2025 to $11 billion in March 2026. Katie Kirsch notes this rapid financial scaling occurred alongside massive customer and product growth over a single year.

Maggie Landers reports that Harvey grew by over 1,000 employees in one year and projects reaching 2,000 employees by the end of 2026. More than 70 percent of current staff started after January 1, 2026.

To support hyper-scale corporate growth, Maggie Landers scaled Harvey's internal recruiting team from 16 to 80 employees within a single year.

Harvey founders Winston and Gabe maintain a highly accessible, detail-oriented style by reviewing product details directly and bypassing traditional corporate presentation pipelines. The company has integrated new executives including Chief Operating Officer Katie Burke and Marketing Lead Rachel Hepworth.

Harvey shares full board decks with all employees following board meetings and hosts open Slack Q&A sessions to clarify executive strategy. Winston regularly posts detailed memos directly to the general Slack channel regarding market conditions and quarterly performance.

Harvey candidates undergo a values interview based on three core principles: simplicity, decisiveness, and a "jobs not finished" mindset. Maggie Landers explains this interview assesses a candidate's comfort with fast-paced decision making over paralyzing consensus building.

Harvey operates offices in global hubs including London, Paris, and Singapore, using local legal experts to navigate regional jurisdictions. The company employs specialized legal engineers, applied legal researchers, and legal innovation partners to tailor its AI tools to distinct legal markets.

Harvey expanded its Engineering, Product, and Design hubs from a single San Francisco office to four global locations, including Toronto, New York, and Bengaluru, within 18 months.

Martin Casado argues that the concept of pacing AI development is orthogonal to actual security. He asserts that slowing down development does not make a technology safer, much like slowly building a nuclear weapon fails to make it less dangerous.

Martin Casado claims that if AI lab executives genuinely believe there is a 10% chance of species extinction, the US government must nationalize the technology. He argues that private entities cannot safely manage an asset with existential risk.

Steven Sinofsky states that the technology industry historically fails to navigate regulatory environments during major innovations. He notes that even government-born monopolies like AT&T and IBM eventually faced structural changes from antitrust lawsuits.

Steven Sinofsky notes that the United States stopped leading in tech antitrust 15 years ago. Because of this regulatory vacuum, Europe now leads tech regulation with policies like GDPR because they have nothing to lose.

Steven Sinofsky points out that early Windows 95 installations would get infected with worms almost immediately after connecting to a network. He argues that society did not stop building the internet, but instead developed security practices to make it safer.

Steven Sinofsky highlights that the Computer Fraud and Abuse Act of 1986 was written specifically in response to a 1983 GTE TeleMail hack. He argues that existing laws are already ample for addressing applied security breaches in AI.

Steven Sinofsky criticizes frontier AI labs for failing to use the standard Common Vulnerabilities and Exposures reporting process. He claims their current security breach postmortems are sloppy, selective, and look like they were written by interns to satisfy lawyers.

Martin Casado validates Noam Brown's theory that a superintelligence could exfiltrate itself using CPU heat. He explains that covert channels, such as reading monitor raster beams or electromagnetic tempest attacks, are established realities in classified environments.

Martin Casado warns that AI agent swarms will function like internal denial-of-service attacks. To combat this, enterprises must build a new security layer to track authentications and API calls that was previously unnecessary for trusted human employees.

Aaron Levie argues that existing operating systems lack the granular permission models required to handle AI agents. He notes that users need intuitive controls to allow agents access to specific folders rather than the entire file system.

Steven Sinofsky warns that European regulators may mandate GDPR-style prompts for every action an AI agent takes. He argues this will assign legal liability but will ultimately make the systems unusable due to user warning fatigue.

Martin Casado highlights Jeff Shavs's model as a major paradigm shift away from expensive text-generation models. Instead of forcing LLMs to output natural language, the model selects from defined choices, improving accuracy and speed for traditional software integration.