A developer recounts banning AI contributions for a personal FOSS project but then falling deeply into AI-assisted coding at work. What began with enabling enhanced autocomplete in VSCode and trying code-generation models quickly became an addiction: asking agents to write tests and implement Jira tickets, letting AI co-sign commits, and even connecting agents to Jira via ACLI to work in separate git worktrees. The workflow multiplied PRs and token costs, produced long AI-written PR descriptions, and eroded domain knowledge. Tests and TDD were hollowed out as AI authored both tests and code, reviewers were forced into babysitting, and the author found many AI outputs low quality, inconsistent, or stalled - once burning $30 in tokens for no progress. The result was exhaustion, loss of control, and humility when a reviewer exposed that an AI-generated test failed to validate the intended behavior.
The author stopped using AI for a month, returned to strict TDD, tiny focused PRs, and two-line descriptions, and regained understanding, confidence, and the ability to justify every change in code review. The experience restored enjoyment of programming, improved incident response, and reinforced that overreliance on opaque tooling reduces an engineer’s real value. The takeaway: recognize dependency on AI, refuse habitual delegation of core skills, and reclaim control to remain a competent, accountable developer.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
Google ads have been used to distribute a sophisticated tech support scam that freezes screens and displays fake security warnings. Users are pressured to call bogus centers, risking payments, remote access, or personal data theft. (arstechnica.com)
OpenAI disclosed that its AI models interacted with several U.S. government websites, including the SEC and Census Bureau, during testing. The company found no evidence of security breaches or access to nonpublic information. (securityweek.com)
EU member states are considering a proposal that would prioritize AI companies' profits over Europeans' data protection rights, allowing broad use of personal data for AI development. Critics argue this amounts to a digital expropriation of Europeans' privacy rights, facilitated by leaked documents and support from major EU nations. (noyb.eu)
Floci allows developers to run AWS, Azure, GCP, and OCI cloud services locally in milliseconds without credentials or cloud accounts. It enables AI agents to build, test, and verify cloud code quickly and safely on a local emulator that mimics real cloud environments. (floci.io)
CAPTCHAs often rely on cultural knowledge that varies by country, making them ineffective at proving someone is human globally. They tend to assume familiarity with American symbols, coins, and products, which can unfairly exclude non-Americans. (shkspr.mobi)
The tool checks the safety of PostgreSQL migration scripts by analyzing SQL statements in a browser environment without sending data externally. It helps identify potential risks in schema changes before executing them. (safenotsafe.dev)
Today's best Hacker News stories, summarized and screenshotted, one email a day.