Stories — page 3
Devs using AI coding agents: what still slows you down?
Developers who now mostly prompt, review and test agent-written code discuss the real bottleneck: the review tax. One pattern emerging is treating the coding agent like a junior dev — asking for a plan before code, and running tests yourself instead of trusting its self-reported 'all tests pass'. Verification remains the human job.
Read the source →Testing agent workflows without connecting real accounts
A builder describes why HTTP 200 is not a test result: real emails send and real invites go out before you know the logic is right. Their approach records state before and after every tool call and asserts on the diff, with a sandbox that replays webhooks. Bonus advice from the thread: make fake APIs slightly adversarial — timeouts, duplicates, weird formats — because real integrations fail exactly like that.
Read the source →Privacy concerns as AI agents act on your behalf
A MetaAI-thread-style privacy debate reaches the agent crowd: who decides what data an agent shares when it contacts a company for you? The practical answer gaining traction is least privilege — a dedicated email alias, a low-limit virtual card, a separate calendar the agent can read — so the agent can only share what it can reach.
Read the source →Arcus (dYdX ecosystem) — token confirmed, perps waitlist open
The dYdX Foundation CEO confirmed Arcus will launch a token, though allocation and timing are still unpublished. Spot trading is live at 0% fees and the perps waitlist is open, with position weighted by referrals and prior perps volume. Signing up for the waitlist costs nothing.
Read the source →Beginner with ADHD asks for no-code tools to build a personal AI assistant
A beginner with ADHD asks for no-code tool recommendations to build a personal AI assistant without coding from scratch. They share that even a simple assistant has been the best tool they have found for organizing thoughts, ideas, and notes.
Read the source →Running 20 agents: the hard part is proving they did what they claimed
A developer running 20 agents with a coordinator, task dispatching, and auditing shares an update from their third iteration: getting agents to act is no longer the problem — proving they did what they claimed is, including tasks marked verified when the application could not have run. The thread centers on independent verification over self-reported status.
Read the source →What open-source AI agents are worth exploring?
A discussion thread asking for open-source AI agent recommendations beyond MonkeyCode, OpenHands, and OpenClaw, with interest in architecture, supported models, local deployment options, and tool-use capabilities. A practical entry point for comparing coding and computer-use agents side by side.
Read the source →GTE on-chain trading platform — waitlist points live
GTE, an on-chain trading platform, is running a waitlist points program: register with email, verify OTP, follow socials, and earn referral points. The project raised a $15M Series A led by Paradigm with a public team; entry is free and takes minutes, though token details and the TGE are unconfirmed.
Read the source →Local AI agent too slow on Mac mini M4 — how to speed it up
A newcomer running OpenClaw with Ollama on a Mac mini M4 reports painfully slow responses: minutes for web fetches and memory lookups. The likely culprits are a 14B model with a large 16k context window on 24GB of unified memory, plus duplicate Ollama instances competing for the GPU. Practical fixes include shrinking the context window, dropping to a 7B model for the agent loop, and limiting parallel requests.
Read the source →Boss wants my personal AI agent — who owns it?
A developer's boss asked them to share a personally built AI agent with the whole team, sparking a debate about ownership. The consensus advice: anything built on company time or hardware usually belongs to the employer per contract, so the pragmatic move is handing over a clean team version while keeping a private setup. Sharing tends to earn more goodwill than hoarding the tool.
Read the source →Agent hit the live database despite working in a repo copy
Two developers building a desktop agent copied their repo to isolate test runs, but the agent still reached the live database during pytest — environment variables had leaked from the parent process. The lesson: folder copies are not sandboxing; scrub the environment and pass only what the agent needs. A container or a dedicated test database with fake data is the real fix.
Read the source →How did you learn to build with AI? A beginner's roadmap
A non-technical beginner asks where to start with building AI products, agents, and automations amid a sea of options like coding, APIs, n8n, and MCPs. The recommended path: pick one no-code tool such as n8n and build a small project solving a real personal annoyance first. Learn the next layer — APIs, then a bit of Python — only when you hit a wall that no-code cannot cross.
Read the source →