Google VP Josh Woodward
Josh Woodward says the iPhone Duo launched the previous Friday and GeminiApp supported it on day one. He emphasizes optimization for the device without detailing the changes in the short post.
OpenAI Codex and ChatGPT team member Thibault Sottiaux
Thibault Sottiaux announces easier switching between multiple accounts on mobile, with desktop support coming soon. He labels it day 26 of his updates. He also teases resets over the following two days and uses the word “banked.” The post provides no reset times, quotas or accumulation rules, so those conditions cannot be inferred.
Former Claude Code team member Thariq
Thariq argues that people are generally poor at predicting how additional intelligence will appear in the economy. He rejects the simple story that it will cost a great deal and make everyone unemployed, worrying that repeatedly telling it could help bring it about. His response is to take a well-aligned ASI more seriously rather than expect less capability. He wants such a system to explain its work, bring people along and make each individual more effective. This is his preferred direction for the technology, not a validated employment forecast.
Replit CEO Amjad Masad
Amjad Masad calls apps hosted on Replit with autofix enabled “Self-Maintaining Software.” He names those two prerequisites without specifying which failures can be fixed automatically. Another post introduces Replit for Government. A third shares a team white paper on sandboxing agents to increase productivity while reducing risk. The updates cover operations, a government offering and execution isolation; the posts do not reproduce the white paper’s design.
Vercel CEO Guillermo Rauch
Guillermo Rauch says the linked tool is improving rapidly, becoming smaller, faster and more token-efficient. He describes continuous benchmarking and asking it to improve its own performance across open and proprietary models. The visible text supplies neither numerical results nor the tool’s name.
Product designer Ryo Lu
Ryo Lu shows ryOS running on the iPhone Duo and describes it as better than iOS, his own assessment. He also shares an entry point for trying it. Readers seeking TestFlight access can message him, the participation method stated in the post. The visible text does not detail ryOS features, and the comparison is not a benchmark.
Y Combinator CEO Garry Tan
Garry Tan positions GBrain Cloud as a fast way to get GBrain working across multiple agents. In his view, harnesses have different strengths: Grok, Muse and Dots for personal use; OpenClaw and Hermes for raw power; Claude Code, Codex and Capy for code generation. GBrain lets context work across those tools. He links an entry point without explaining migration steps in this post. These are his judgments about tool roles, not independent capability rankings.
Builder Zara Zhang
Zara Zhang publishes a video essay about creating tasteful work with AI without being a designer. She adds that a coding agent edited the video from a single, very long prompt. This extends the coding-agent use case into video production. The posts do not include the full prompt or production steps, so this is not a reproduced tutorial.
OpenClaw builder Peter Steinberger
Peter Steinberger says unpredictable inference quality was a common reason he advised bypassing intermediaries. He welcomes a linked change, although the visible text does not identify the service or default setting. Another post argues for using one’s own harness rather than letting companies change permitted actions at will. He also supports Andy Yen and Proton, sympathizing with engineers who feel a loss of identity as AI changes their work. At the same time, he understands that a business operator may have limited patience after two years to learn and adapt. His posts connect control over tools with the human process of adjustment.
SPC General Partner and Bevel Health co-founder Aditya Agarwal
Aditya Agarwal argues that investors and the press fixate too much on the last private-round price. He uses hypothetical ANT IPO valuations of $1.2 trillion or $2 trillion to say that either would represent impressive value creation within six years. He also argues that most company value is created after listing. This is his valuation perspective; the hypothetical figures are not verified listing plans or prices.
Anthropic AI assistant Claude
Claude’s official account announces that Max 5x is being renamed Max and Max 20x is becoming Ultra. Max remains $100 per month and Ultra $200, with no action needed from subscribers. The five-hour limit is being removed for verified Ultra accounts, starting that day with most existing subscribers. Newer accounts retain a temporary limit until verification, which the account says helps prevent fraud and abuse. Subscribers will receive an email when the limit is removed. Weekly limits remain unchanged, so removing the five-hour limit does not mean unlimited usage.
No Priors — Taking on ServiceNow: Coding Agents the Rest of Us with Serval Co-Founder and CEO Jake Stauch
Core takeaway: Enterprise agents create value not only by writing code, but by making each privilege escalation, approval and execution traceable.
Serval co-founder and CEO Jake Stauch previously spent years serving IT teams at Verkada. He sees an opportunity to bring coding agents to IT departments whose primary job is not software development, then extend them to HR, finance and legal. Serval chose to challenge ServiceNow directly: drag-and-drop automation remains complex for nontechnical users and awkwardly abstract for technical ones. Code can express workflows, while AI lowers the barrier to writing it.
“The boundaries are the product for us,” Stauch says. Employees at large enterprises have limited permissions, which also limits their personal agents. Serval combines service-account access connected by IT with deterministic approval workflows, tickets and audit records, permitting temporary access to run a specific workflow once. As capabilities increase, excessive permissions can cause greater harm. He therefore expects individual permissions to become narrower.
A second principle separates development, staging and production. AI-generated password resets and onboarding or offboarding automations still need approvals, user acceptance tests and release gates. Background agents can propose tools or scripts and obtain approval incrementally. Stauch acknowledges that this still requires supervision and does not fully solve the tension between autonomy and enterprise confidence.
Deployment still needs people. The company sends deployment strategists and software engineers on-site to understand customer processes, with its ProServal automation-engineering team supporting subsequent demand. Capabilities customers repeatedly build themselves may become platform features, such as identifying conflicting or outdated knowledge articles and reclaiming unused software licenses. Administrators can also talk to Catalyst to turn requirements into automations, approval rules and permission checks without first learning every product primitive.
Finally, token spending is not work value. Stauch says customers want ROI evidence and the team tries pricing by workflows run, while the industry has not fully solved how to measure what tokens buy. He argues that rebuilding core systems of record still incurs maintenance costs: smaller point solutions are more likely to consolidate, while the core categories persist. He reports growing from 20 staff in January to roughly 200 and links higher individual productivity to greater hiring demand. These are company accounts and judgments. Enabling more builders still leaves the organization responsible for production releases.