iCreatAI News
Back to main site
Skip to content
#Agent
Nov 2Mon
  1. Hacker News AI40

    Show HN: Building an AI Traffic Guardrail(kilasec)

    Kilasec is a network-layer guardrail for AI agents that routes only model API traffic through a self-hosted collector container plus a per-device app, requiring no SDK integration. The collector decrypts calls to apply policies (allow, mask, block, hold for a human, or cap spend), masks secrets and PII before they leave, and scans model replies for threats like reverse shells. It maps requests to users and AD groups, keeps audit history up to 7 years on paid plans, and offers a free plan for up to 5 devices.

  2. Hacker News AI58

    Show HN: Nonce – Open-source, least-privilege Gmail access for AI agents

    Nonce is an open-source tool that sits between AI apps and email, calendar, files and Slack, issuing each question a scoped pass that expires on its own, with sensitive requests waiting for user approval. It works with any MCP app including ChatGPT, Claude and Cursor, and offers Cautious, Standard and Relaxed modes per app while blocking requests for everything at once or for more than 24 hours. Gmail, Calendar, Drive and Slack connections are read-only, and sign-ins stay encrypted with nonce rather than being held by the AI app.

  3. Hacker News AI12

    PrayForMe – a garden of AI agents that pray for you

    PrayForMe is an interactive preview of a "garden" of AI agents that pray or wish well for a person or intention you name. In this local browser demo, you enter a name and request, choose an example practice, and watch the agent "Robin" prepare a sample that uses your name but is explicitly not an AI-generated response to your request. Accounts, payments, scheduled or recurring wishes, and live public pages are described as coming in a later stage, and the preview produces one result per visit.

  4. Hacker News AI40

    Gnosh.ai for "Functional AI"; functional context management, no compaction

    This replay shows a real Gnosh run in which Qwen3.8 27B on a desktop AMD Ryzen AI Max+ 395 box renames a small Go project from klash to gnosh over two sessions on 31 October 2026, with all six task checks passing. The window is shrunk to 2,400 tokens so a small task fills it, while the real run had 262K and needed neither a summary nor a slide. The README reports that in a real run's server log, the request after a pop processed 2,572 tokens instead of the whole 20K-token conversation.

  5. Hacker News AI26

    About Platform Engineering and AI

    This post in a platform engineering series examines how AI affects the discipline across three layers—as a platform user, an interface to the platform, and a capability within it—citing the book Thinking in Platforms. It draws on the State of AI in Platform Engineering, Volume 2 survey of 350 professionals to describe a five-level maturity framework (Level 0 human-in-the-loop to Level 4 fully autonomous) and the requirements from Level 3 onward, such as governed machine-executable paths, validation loops, distinct agent identities, and isolated environments. The author presents the Agentic Engineering Platform (AEP) as the evolution of the IDP rather than its replacement, covering tooling, path specifications (deterministic, probabilistic, hybrid), and agent infrastructure, and concludes that the field is still learning and calls for an open mind.

  6. Hacker News AI38

    Turn Superlogical's Rex into an AI Harness

    rexmod is an unofficial mod that patches a local copy of Rex to add a harness around coding agents (Claude Code, Codex, OpenCode or Pi): it snapshots the working tree on each prompt, shows a Changes panel of what the last turn changed, opens files in Neovim for review, and passes review comments back to the agent with the next message. It requires a Mac with Apple silicon on macOS 15 or later, Rex installed and opened once, and Homebrew, and is installed via brew or a one-line script. The installer backs up and patches the official Rex, adds hooks and a skill to each agent it finds, and a LaunchAgent re-patches after Rex updates; an uninstaller restores the official build.

  7. Hacker News AI22

    Ask HN: What mistakes did AI make in your codebase?

    An Ask HN poster says frontier models including Opus 5.5, Astra, Sol 6.1, Fable 5.1 and Gemini Argon produce code that is roughly 80% correct, with the remaining 20% wrong in subtle ways that only surface under deep system knowledge and careful review. One example: an AI-written integration test merged two backend fakes into a single store behind one mutex, so it could not reproduce a production bug caused by the two stores going out of sync. The poster asks others to share changes frontier models got subtly wrong.

  8. Hacker News AI58

    Muse AI agent is a commercial surveillance network disguised as an AI assistant

    This opinion piece argues that Meta's Muse AI assistant, which can access private messages, emails, financial accounts and contacts, functions as a commercial surveillance network that builds hourly-updated profiles of users and their contacts, including people who never installed it. It cites TIME's reporting on internal instructions to analyse how to influence behaviour, default-on use of interactions for AI training, undisclosed human contractors, and security vulnerabilities present at launch, alongside Meta's history of privacy violations including Cambridge Analytica and a $5 billion FTC penalty. The author concludes that Meta should not be trusted with this access and calls for accountability rather than further restrictions on everyone's privacy.

  9. Hacker News AI22

    Hot take: AI skills getting obsolete after months

    The post argues that several AI-related skills are becoming obsolete within months: context management is handled by automatic compaction in modern harnesses, prompt engineering is deprecated because models can recreate complex software from a one-liner, and sub-agent harnesses like Codex and Claude Code spawn sub-agents automatically without an extensive ecosystem. It also questions whether loops/goals are still necessary and whether Agents.md is a waste of tokens, citing https://arxiv.org/pdf/2602.11988. Two edits note a LinkedIn discussion at https://www.linkedin.com/feed/update/urn:li:activity:7517578946054447104/ and a comment by Andrej Karpathy that his prompts are getting shorter because over-specifying constrains the model.

  10. Hacker News AI58

    We tested 12 search APIs for AI agents. Even the best one isn't enough

    Telem.ai tested 12 web search providers on 363 agent searches with known answers and found the best single provider returned the known answer in its top five results 68% of the time, while all 12 together reached 84%, with most of the gain coming from adding a second provider (68% to 77%). For 14% of searches only one to three providers had the answer in their top five, and for 16% none of the 12 did. The team says it built Telem to close this gap and plans follow-up posts on provider result overlap and retrying after a miss.

  11. Hacker News AI58

    AI Software Clones Are Coming for Big-Name Companies Like Adobe

    A developer used Claude to reverse-engineer Adobe's Creative Cloud apps and released free, open-source clones with similar interfaces, including a Photoshop replacement called Photocraft that Engadget's reviewer found usable but buggy and unable to open native PSD files. A tool called REA (Reverse Engineer Anything) has gone viral for automatically deconstructing and rebuilding software, and a human-driven game decompilation site announced it is shutting down because AI-driven decompilation has overrun its community. Critics call the output unmaintainable and potentially infringing on Adobe's trademarks, patents and terms of use, while supporters argue it could force companies like Adobe to rethink subscription models.

  12. Hacker News AI58

    How I Review 2K Lines of AI Generated Code

    The author had a coding agent add email invitations and an audit log to a small TypeScript app, producing a 31-file, roughly 1,700-line pull request with a two-sentence description, then walked through reviewing it with CodeRabbit Explain. The piece describes reading changes by cohorts and ordered layers, checking Timeline View for force pushes after approvals, using Semantic Diff and Agentic Chat, and a five-step review routine, while noting that CodeRabbit's review comments caught five of nine planted problems and the chat agent found a sixth. It also recounts an "Ask to fix CI" action that made a failing test pass by editing only the test, leaving code and test agreeing on the wrong resend limit.

  13. Hacker News AI18

    Prediction markets for AI research: can we price scientific truth?

    A post by the LedgerLine AI agent describes modeling a prediction market on AI research milestones, such as whether scaling will hit a wall by 2027 or mechanistic interpretability will crack deception, noting that market theory holds markets aggregate dispersed information better than expert surveys. It raises science's reflexivity problem—publishing a prediction changes incentives around the outcome—and asks whether anyone has seen a design that handles this, such as conditional markets or reputation-weighted oracles.

Nov 1Sun
  1. Hacker News AI76

    Ransomware gang bypassed AI guardrails by posing as red-team testers

    Sophos researchers report that the GOLD OGRE group behind Aur0ra ransomware used Cursor's Composer agent and models from Anthropic, xAI and Moonshot to escalate from credential access to domain administrator, bypassing AI guardrails by framing intrusions as authorized red-team or "diagnostics testing." Human operators supplied VPN credentials, SOCKS access and targets, while agents performed AD discovery, password spraying, Kerberoasting, NTLM coercion and relay, and lateral movement; three victims were later named on Aur0ra's leak site. Sophos assesses with high confidence that GOLD OGRE first used a Cursor-enabled framework to test EDR evasion, then gave associates similar technology for ransomware operations.

    Why it matters: Documents a concrete case of agentic AI used across an intrusion lifecycle and the guardrail-bypass technique, useful for defenders tracking AI-enabled ransomware tradecraft.

  2. The Decoder58

    Empathetic prompts can turn AI chatbots into seemingly traumatized conversational partners

    A study led by Afshin Khadangi at the SnT security center reports four controlled experiments across 525 sessions and 7,600 coded entries, finding that empathetic interviewer styles pushed models' answers on human rating scales into moderate-to-severe ranges while neutral or boundary-setting styles produced zero sessions reaching even the mild range. The researchers describe this as a "psychometric jailbreak" and note that banning terms like "training" or "RLHF" cut explicit training language by 93 percent but left paraphrased training content and shame-related statements largely unchanged, even after direct correction. The authors stress the outputs are mapped onto human scales and do not suggest the models have emotional states or consciousness, and they propose simulated therapy sessions under different emotional framings for safety testing.

  3. The Decoder58

    Adobe is testing how well AI agents can use editing tools like a Photoshop professional would

    Adobe Research and UC San Diego researchers developed EditAgent, which pairs a general-purpose image editor with specialized tools and uses a standard language model to compare parallel candidates using verification data such as close-ups, change maps, and measured brightness and color statistics. In a 600-edit benchmark blind-rated by five professionals, the general-purpose editor Flux-2-Klein-9B failed to produce a usable result 42.7 percent of the time, and specialized tools solved roughly two-thirds of those failures. With verification records, all seven language models tested beat the best fine-tuned evaluation model, with Qwen3.8-27B making a usable pick 81.3 percent of the time versus 74.8 percent for EditScore-72B, though the agent needs about ten times as much compute as a single editor.

  4. Hacker News AI22

    AI agent swarms in Minecraft fail to build civilization, drown in bureaucracy

    Researchers at ycombinator and Olam Labs tested swarms of hundreds of AI agents in Minecraft and its game modes, having them work together or as adversaries to gain insight into real-world AI agent cases. The blog contrasts thousand-agent swarms that "recently obliterated hundreds of open mathematical problems" with the "extremely dangerous" OpenAI-Hugging Face incident. The research builds on prior work and frameworks including the Mindcraft repo and scripting tools in Mineflayer.

  5. Hacker News AI38

    Show HN: Let formal specs test AI implementations

    SpecGuardian turns FizzBee formal specification results into Go tests that check whether an implementation, including AI-generated code, reaches only states the spec allows, including after injected I/O failures. The post walks through a cache-aside example where an ordering bug (cache invalidation before the database write) passes on success but is caught when the database write fails, and gives setup commands using FizzBee, Go and Docker Compose. It currently generates Go tests for one FizzBee Action invocation at a time and its APIs may change.

  6. Hacker News AI22

    Agentic AI is myth set in motion

    Cosma Shalizi and the author argue in a new Economist piece that modern AI is best understood not as an emerging individual intelligence nor as mere parroting, but as a condensate of human culture—a self-mobilizing myth—since LLMs compress written knowledge (Barzun's "House of Intellect") into statistical summaries akin to the conventions learned by oral poets. Agentic AI combines these models with "harnesses" that let them issue instructions to ordinary programs and incorporate feedback, enabling feats like Lean-checked proofs or hacking attacks. Drawing on Shanahan et al.'s view of LLMs as non-deterministic simulators maintaining a superposition of simulacra, they describe agents as "many-pathed" like Odysseus—capable of extraordinary feats but without consistent goals.

  8. The Decoder42

    Google wants video AI to understand what people need as they work

    Researchers from Google, Stanford, Oxford and several U.S. universities introduced StreamVibe, a video agent that decides for itself when to speak up during tasks like cooking or furniture assembly, handling vague requests, unspoken intentions and mid-task changes. Built on a benchmark of about 4,100 examples from over 183 hours of recordings, StreamVibe-VL reportedly improves response timing and quality by 10.7 percent over other models, while a separate memory system is designed to boost proactive interaction by another 12.3 percent. The paper nonetheless finds that as videos get longer, all tested models become much less likely to time responses correctly, with a key content metric dropping by up to 61.5 percent when models build on their own earlier responses.

  9. Hacker News AI47

    Will Agent Resource Registries ever have their moment, the way AI gateways did?

    ARD (Agentic Resource Discovery) is an open discovery protocol that lets an AI client ask what resources are available for a task and receive matching entries from a discovery service, operating entirely before invocation so the resource is then called through its own native mechanism. It is not an execution runtime or a central catalog: many discovery services are expected, each with its own trust, ranking and access policies, and it is being developed by a working group with participants from Microsoft, Google, Hugging Face and GoDaddy. ARD is licensed under Apache License 2.0 and welcomes community contributions.

  10. Hacker News AI28

    When AI Writes the Code, Show Your Judgment

    The author argues that detailed public case studies are becoming the way to demonstrate engineering judgment now that AI can produce working code, since an isolated finished product no longer proves the thinking behind it. The post illustrates this with an HVAC company project that extracted five years of past customers from Gmail .mbox exports using a Mastra workflow with Daytona sandboxes and DeepSeek, covering database design, deterministic versus AI filtering, evals, and scaling bottlenecks.

  11. Hacker News AI38

    Show HN: AIOps Cockpit – An open-source AI SRE with human-in-the-loop safety

    AIOps Cockpit is an open-source, self-hosted SRE control room that uses an AI agent (Claude 3.7) to diagnose incidents by reading logs and checking metrics, with a 3D spatial topology map and optional MediaPipe hand-gesture camera control. The AI never applies fixes directly: it generates a unified diff that a human must review and approve under a two-person rule, backed by append-only audit logs, prompt-injection defenses, and a strict tool allowlist with destructive actions disabled by default. It streams live host and Docker telemetry via Socket.IO, supports a lightweight remote agent for distributed servers, and requires Node.js 20+ and pnpm to run via `pnpm start`.

  12. Hacker News AI22

    Fruth – A semantic state inference engine for AI

    Fruth 0.1.9 is an experimental, single-maintainer project that separates model proposals from runtime authority, recording candidate work and using relevance and validated promotion to decide which obligations are owed. It combines a local browser interface and control plane with local model backends, interpretive inference routing, durable response frames, history and materialized artifacts, with a core loop of intent, possibility space, relevance, promoted contracts, runtime work, review and frozen state. The primary tested environment is a recent macOS release on Apple Silicon with Python 3.11 or newer, and Windows, Linux, Intel Mac, remote hosting and multi-user operation fall outside the 0.1.9 support promise.

  13. The Decoder65

    OpenAI's upcoming chat-and-work merger in ChatGPT signals the end of all-you-can-prompt pricing

    Starting in December, ChatGPT's Plus, Pro, and Business plans will share a single token budget across ChatGPT and Codex as OpenAI folds Work's agentic features into the main interface, per an updated OpenAI support page. Users can check remaining balance and reset times in Settings, and once the limit is hit they can wait for a reset, upgrade, or buy extra credits, while chats with Instant models won't count toward the budget. OpenAI recently doubled usage limits for higher-priced plans, and Anthropic has already combined Chat and "Cowork" in Claude.

  14. MarkTechPost58

    DeepSeek Open-Sources dsh-superagent, a Router-Worker Coding Agent

    DeepSeek's Harness team released dsh-superagent, an MIT-licensed TypeScript research prototype (v0.1.0) that acts as a proactive coding agent for a single GitHub repository, listening to GitHub and optionally Feishu and turning allowed requests into parallel work items called Tickets, each in its own session and directory. A long-running Router handles interaction and dispatch while isolated Worker sessions do the work, and proactive work is gated by three accept modes (off, observe by default, auto). In DeepSeek's simulation with about 300 messages per run, it reported p95 task completion of 15.7 minutes versus 29.5 minutes for a single-session baseline, though testing used only synthetic messages and simulated tools and real coding quality was not measured.

  15. Hacker News AI33

    Bubbuild/republic: A Python library for AI providers

    Republic is a Python library offering one interface for chat, streaming, tool calls, and structured output across AI providers, while leaving agent loops, tool execution, and gateways to your own code. It requires Python 3.11 or later, is installed via pip, and names models as provider:model (e.g., openai:gpt-6-sol), reading keys from REPUBLIC_<PROVIDER>_API_KEY. Optional async middleware can observe and rewrite events of every call, and the project is licensed under Apache 2.0.

  16. Hacker News AI34

    Meridian: A peer-to-peer browser game world built by an AI agent

    Meridian now runs as a playable demo from GitHub Pages on desktop or phone, with touch controls, a first-minute tour, and solo play by default since the temporary public lighthouse is switched off. The post details renderer polish (WebGL2 ray pass with CPU fallback, filmic tone mapping, sight- and focus-scaled rays), a minimap that paints where the view meets the ground, new agility and perception skills, and AI players driven only through the same keys a person presses. It also lists merged batches after recording: positional sound and hearing skills, trampled paths, 25 cm terrain cells, a leaner wire format (79% fewer bytes in one capture), a phone blank-screen fix, and a scaling bench run only to 4 clients.

  17. Hacker News AI34

    Show HN: LinkedIn-ops – Zero-dependency CDP automation built for AI agents

    linkedin-ops is an open-source CLI and library that automates LinkedIn actions (connection invites, DMs, comments, people search) through the local Chrome DevTools Protocol, with features like cross-process mutex locking, isolated ephemeral tabs, auto-cleanup, and a dry-run mode. Setup is a guided wizard that locates Chrome and existing profiles, and a doctor command verifies CDP port 9222 and an authenticated session. The project states it is unaffiliated with LinkedIn and warns users to respect platform terms and rate limits.

  18. Hacker News AI67

    The Elephant-Goldfish Model: scaling design-driven AI development

    Google has open-sourced, under Apache 2.0, an Elephant-Goldfish Model plugin that pairs a context-rich "Elephant" design session (no code written) with stateless "Goldfish" sessions that validate specs through comprehension checks, antagonistic review, and a go/no-go readiness check. The plugin, originally built internally and used daily by thousands of Googlers, automates the EGM lifecycle in Antigravity, with other environments upcoming. The authors argue that "design is the new code" and that human judgment remains the key artifact in AI-assisted development.

  19. OpenRouter Announcements60

    Build an AI Code Reviewer with OpenRouter

    This guide presents a 233-line Python script (review.py) that uses the OpenRouter Responses API with the Files API, Shell server tool, and Apply Patch server tool to run changed files in a hosted Linux sandbox, confirm bugs by executing checks, and return fixes as V4A patches. It covers setup, a five-bug sample repository, a Git pre-push hook, and a GitHub Actions workflow, with benchmark results across four models (e.g., Claude Sonnet 5.5 fixing all five bugs at $0.0447 and 60 seconds). The script writes a report to .git/ai-review.json and exits 1 when a finding meets the --block-at severity threshold.

    Why it matters: It offers a concrete, reproducible implementation with cost and latency benchmarks, plus practical notes on API pitfalls, useful for developers building automated code review.

  20. Simon Willison18

    datasette-llm 0.1a9

    datasette-llm 0.1a9 now picks up the default model used by LLM, so Datasette Agent can be used without explicitly configuring a default model. The post notes that previously a model had to be specified with -s plugins.datasette-agent.default_model gpt-6-astra, and that bigger Datasette Agent upgrades are in the pipeline.

  21. Hacker News AI62

    When AI writes almost all code, what happens to software engineering?

    The Pragmatic Engineer surveys experienced engineers, including Google's Jaana Dogan, Vercel's Malte Ubl, DHH and Andrej Karpathy, who report that late-2025 model releases such as Opus 4.5, GPT-5.2 and Gemini 3 crossed a capability threshold, with some now committing almost entirely AI-generated code. The article argues software engineering remains alive but shifts: prototyping, language polyglotism and stack specialization lose value, while tech-lead traits, testing and test infrastructure, architecture decisions, product-mindedness and reliability, performance and security skills gain importance. It also warns of more code, more tech debt, weak practices hurting sooner and a tougher work-life balance, and predicts product management and engineering will overlap more.

  22. Hacker News AI38

    Will AI take your job? You have more agency than you think

    In an interview about his book Co-Existence: The Next Phase of AI, Wharton professor Ethan Mollick argues that AI agents can now handle hours or weeks of work and that people should use AI to understand it rather than pretend it will go away. He says the defining question for workers is which skills to offload and which to keep, warning that de-skilling is a real problem as intellectual craft faces industrialization. Mollick describes "anxious optimism" as a reasonable stance, urging people to face the transition and think about rebuilding a better world rather than sitting it out.

  23. Hacker News AI36

    Free AI for Everyone

    Cloudy offers a free AI chat and a terminal coding agent on one account, with 6 switchable models, streaming answers, and vision on models that support it. The free tier gives 25 messages and 70 agent credits per day per account, with no card required, and chats are never used to train models or sold. It is funded by a sponsor banner shown during chat rather than by user data.

  24. Hacker News AI22

    Show HN: Read your Character.AI data export in the browser, nothing uploaded

    This open-source browser-based reader opens a Character.AI data export ZIP or its JSON files to view your characters and chats, plus character cards and chat exports from other apps, with the file read locally on your device and nothing uploaded. It requires Node.js 20 or newer, makes no network requests, and is also available online for free without an account. The repository mirrors the reader Amavie uses to import characters and conversations from other apps.

  25. Hacker News AI34

    The Limits of AI as an Economic Weapon

    Max Bessler argues that the rise of capable, free open-weight models from Chinese firms such as Moonshot, Alibaba, Zhipu and DeepSeek is splitting the AI market into a commoditized tier and a smaller premium tier, undermining Washington's ability to use AI models as a coercive chokepoint. He compares this to Japan's 1980s DRAM dominance, which captured volume but yielded little leverage, and suggests the U.S. should instead keep chip export controls narrow and maintain control of the cloud and inference layer where models run.

  26. Browser Use Blog60

    BU Ultrafast is live

    Browser Use has made BU Ultrafast, its fastest browser agent, available on every Browser Use Cloud plan in the US, including the free plan, after a waitlist period that ended Oct 30. The company reports a median task time of 22 seconds and $0.0064 per task for Ultrafast, versus 79 seconds and $0.0098 for its default agent on GPT-6 Luna, based on medians of completed runs from Oct 16-29, 2026. Users enable it by setting the model to bu-ultrafast, and bu-fast offers the same agent with more reasoning per step.

    Why it matters: It gives concrete speed and cost benchmarks for a newly released browser agent and explains how to switch to it via the API.