Developer Builds Physical Gear Shifter to Switch Between Claude AI Models

·1 views

A developer shared on X a working physical car-style gear shifter that switches between different Claude AI models — assigning Haiku, Sonnet, and Opus to separate gear positions. The device was built using Claude itself, creating a feedback loop where the model helped reduce friction in its own use. Under the hood, a small script translates the shifter's position into API calls that update which Claude model is used for the next request. The project highlights a real productivity issue for heavy AI users: repeatedly choosing between models is a small but frequent mental cost that adds up over a workday. The build demonstrates how physical, intuitive controls can replace menu-driven decisions, making habitual choices faster and more deliberate.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Nylas Agent Accounts let schools automate admissions emails without human inbox access

Schools using admissions email inboxes often face repetitive queries about deadlines, missing documents, and application status that require manual lookup and response. A new approach using Nylas Agent Accounts allows developers to provision a dedicated, application-owned inbox — not tied to any human's mailbox — that can send, receive, and manage email autonomously. The agent inbox is created via a single API call or CLI command, with no OAuth flow required, and supports the same endpoints as standard connected accounts. It can handle tasks like downloading PDF attachments, tracking applicant state, and sending deadline reminders without human intervention. Free-tier limits include 200 messages per account per day and 30-day inbox retention, with new sending domains requiring roughly four weeks to warm up with major email providers.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

JavaScript Object Constructor Functions: A Practical Overview

JavaScript object constructor functions allow developers to create reusable object types by defining properties using the 'this' keyword. By convention, constructor function names begin with an uppercase letter to distinguish them from regular functions. New object instances are created using the 'new' keyword, which binds the defined properties to each instance. Additional properties can be added to all instances of a constructor via the prototype, while individual methods can be attached directly to specific objects. This pattern enables flexible and structured object creation in JavaScript applications.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Solana Token Extensions Let Issuers Mint and Revoke Credentials Without User Consent

A developer building on Solana's Token-2022 program combined the Non-Transferable and Permanent Delegate extensions to create soulbound credentials that holders cannot sell, trade, or transfer. The Non-Transferable extension enforces this restriction at the protocol level, meaning transfer attempts are rejected by the blockchain itself rather than by any application backend. The Permanent Delegate extension allows the token issuer to burn or reclaim a token from any holder's wallet without requiring the holder's signature or approval. This pattern is already in production use: Paxos applies the Permanent Delegate extension to its USDP stablecoin to claw back funds linked to illegal activity under regulatory requirements. The setup raises questions about the nature of digital ownership when an issuer retains unilateral revocation rights enforced entirely by on-chain code.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Darkmoon: Open-Source AI Agent Automates Active Directory Penetration Testing

Darkmoon is a newly open-sourced (GPLv3) autonomous pentesting tool designed to tackle Active Directory attack paths, an area where most AI security tools fall short. The tool uses an agent loop that enumerates, reasons, and pivots across multi-step AD attack chains while maintaining state throughout the process. It follows a readable Markdown playbook and uses BloodHound-style logic to identify and execute attack paths with real tools, logging every step and its output. Because both the methodology and execution are fully transparent, security reviewers can trace exactly how a domain was compromised. Built by pentesters and released publicly, the project is available on GitHub with documentation and a demo video.

0 comments Read more at DEV Community