PhoneClaw, Claude Code show speed gains

A new class of software is moving from autocomplete to delegation: PhoneClaw runs tasks on an iPhone, and Claude Code runs them in a developer’s terminal. (github.com) (anthropic.com) PhoneClaw is an open-source project that says it runs Google’s Gemma 4 model locally on an iPhone, without sending chats or images to the cloud. Its GitHub repository showed 645 stars and 97 forks on April 27. (github.com) The project’s recent updates included an English live mode release and a CPU decoding change that the maintainer described as roughly 1.5-times to 2-times faster. The repo also lists built-in skills for calendars, reminders, contacts, clipboard access, and device information. (github.com) Claude Code takes the same agent idea to software development. Anthropic says the tool reads a codebase, edits files across projects, runs tests, uses command-line tools, and can work inside the terminal, an integrated development environment, or GitHub. (anthropic.com) (github.com) Anthropic is also attaching hard numbers to the pitch. On its product page, the company says Stripe deployed Claude Code to 1,370 engineers, one team finished a 10,000-line Scala-to-Java migration in four days, and Ramp cut incident investigation time by 80%. (anthropic.com) The same language is now spreading into chip design, where engineers spend large amounts of time writing testbenches, debugging waveforms, and rerunning verification jobs. Those jobs are repetitive enough for agents to take over pieces of the loop. (chipagents.ai) (cadence.com) ChipAgents says its platform lets teams “iterate on your chip design and verification 10x faster” inside existing code editors. At DVCon 2026, the company said it introduced root-cause analysis with “multi-agent AI teams” for debugging work. (chipagents.ai) Cadence, one of the largest electronic design automation vendors, made the market signal clearer in February. It launched ChipStack AI Super Agent and said the product can deliver up to 10-times productivity gains in design coding, testbench creation, test planning, regression orchestration, debugging, and automatic fixes. (cadence.com) What ties these products together is not one model or one app store listing. It is a shift toward agents that can inspect context, choose tools, execute steps, and hand back a finished result instead of a suggestion. (anthropic.com) (github.com) (cadence.com) The claims are still uneven: PhoneClaw’s traction is visible in open-source activity, while Claude Code and chip-design vendors are publishing customer and product numbers from their own platforms. But the common sales pitch is now specific, measurable, and spreading across phones, software teams, and semiconductor labs. (github.com) (anthropic.com) (chipagents.ai)