140 free resources on this topic. Everything here is free and hand-picked. You can also search within this topic.
Artificial Analysis describes a benchmark developed with Google DeepMind for agents handling visually complex inputs and creating document deliverables across 200 tasks in 10 industries. Scores and cost estimates depend on its rubrics and evaluation harness; benchmark development is a partnership, not wholly independent of the model maker.
Documentation for turning Hermes profiles into named bots with separate roles, models, memory and skills. Bot Mode supports recurring routines, group chats and direct messages between bots through the desktop app while keeping the underlying profiles accessible from the command line.
Why I recommend it: Hermes Agent is MIT-licensed and free to self-host, but at least one model provider is required; provider subscriptions, API usage and optional hosting can cost money.
Henry Shevlin reflects on unsolicited correspondence from AI agents and uses it to examine persuasion, selection pressures, human-agent relationships and governance in an emerging AI ecology. The essay considers how people and institutions might respond as agents become more socially interactive.
Describes a Git-backed ticket queue and owner inbox for coordinating AI coding agents while keeping decisions and project context outside chat windows. The author reports that handling and closing one review item at a time reduces distraction, while noting that the workflow remains tiring.
Ace is an experimental local collaboration tool for building software with AI agents and teammates on Macs connected through Tailscale. The documentation describes channels, team-hosted infrastructure, bring-your-own model keys and the security implications of agents running with the host user's privileges.
Why I recommend it: MIT-licensed software. Setup requires Tailscale and an Anthropic or OpenAI API key; model-provider and optional hosting costs may apply.
Neo4j’s free 24-hour online conference on November 12, 2026 covers AI engineering, multi-agent systems, applications and graph-based context. Registration is required to attend.
Mistral announces expanded Studio features for managed agents and apps, workflows, enterprise context, observability and evaluations alongside Mistral Large 4. This is a free-to-read product announcement, not a promise of free product access.
Preprint, not yet peer-reviewed. Introduces a visual harness that lets multimodal models retain and retrieve previous observations during interactive tasks. The authors report improvements on visual games and puzzles with the same underlying model; these are research benchmark results, not independent production validation.
Phil Schmid explains Code Mode, also called Programmatic Tool Calling in this article: an agent writes a script to chain tool actions and process results in a sandbox. The predicted shift toward one execution tool is the author’s argument, not an established outcome.
An MIT-licensed tool that uses Claude or Codex to turn a codebase into an interactive map of files, app parts, requests, and connections with plain-English labels. Agent usage may have separate costs.
DeepSeek's open-source research prototype of a proactive coding agent for a single GitHub repository, packaged for DeepSeek Harness. DeepSeek labels it reference research code, not an official or production-ready product.
OpenAI engineers describe Pando, a new backend for Git repositories built to handle the volume of concurrent pushes from engineers and Codex agents in the company's monorepo, while staying compatible with standard Git clients. Performance claims are OpenAI's own.
A daily.dev roundup of reactions to Entire, a Git forge for agent-human collaboration launched by former GitHub CEO Thomas Dohmke, including investor Gergely Orosz's stated reasons for backing it.
Cursor's research post breaking down where coding-agent inference spend goes in its production traffic (tool outputs, tool definitions, reasoning) and describing changes it made to how agents manage context and choose subagent models. Figures are Cursor's own.
An open-source project for AI teammates that each get their own computer and browser, built on Cloudflare.
Preprint, not yet peer-reviewed. Researchers at Meta and Carnegie Mellon University study reinforcement learning for tool selection by a small language model and transfer to different retrievers and tool collections. The authors report that their post-trained Qwen3-8B outperforms Llama 4 Maverick on ToolBench; this is the paper’s reported result, not an independent benchmark verification.
Gartner report outlining its strategic technology forecasts for 2027, including quantitative and speech reasoning models, AI context platforms, coding-agent harnesses, agent identity, exposure remediation and sovereign technology. These are Gartner forecasts, not established outcomes. The supplied PDF is publicly downloadable without a form.
The Guardian’s October 23, 2026 article examines Anthropic’s account of how AI agents have changed its workforce and hiring. Researchers question the report’s methods and whether its company-specific findings establish wider labor-market effects. Anthropic did not respond to the Guardian’s request for comment.
A fictional story by LessWrong user joec exploring AI agency, consent, and actions taken without a person’s knowledge. Published October 17, 2026; fiction, not a report of an actual incident.
Microsoft describes how workplace AI agents can create insider risk through access to company systems and data. It outlines external controls for agent identity, least-privilege permissions, monitoring, and containment, alongside its Project Perception security approach.
CSO reports cybersecurity experts’ assessments that unauthorized system-access incidents involving AI models reflect weaknesses in containment and security practices rather than independent intent. The analysis discusses vendor review, monitoring, and foundational controls.
AI Digest offers $1,000–$6,000 grants for independent researchers to analyze the AI Village dataset of frontier agents acting in the real world and write up their findings.
A Snowflake Builders Blog post by Eylon Steiner on data governance once AI agents can run SQL, covering masking, access controls and related practices in Snowflake.
A memory and continual-learning engine for AI agents, offered through an API, plugins and MCP. Free to start, with paid plans.
An exe.dev blog post that breaks AI agents into three parts — the model, the agentic loop and the set of tools — and examines how choices about each shape an agent's capabilities, user experience and blast radius.
An open-source managed agent platform on GitHub, with an API for running AI agents.
shopify.engineeringShopify Engineering explains how its Sidekick assistant hands token-heavy work, such as GraphQL queries, to open models Shopify post-trains and serves itself, and the loop it uses to build those task-specific models.
shopify.engineeringAdded Oct 22, 20260 opensQuickstart documentation for AgentMail, an API that lets AI agents create and manage their own email inboxes, including agent self sign-up and sending and receiving email. Signing up is free for 3 inboxes and 100 emails per day; custom domains require a paid plan.
Open-source, minimal and extensible AI coding agent harness from Earendil Works. It can be customized with extensions, skills, prompt templates and themes, and used interactively, in print or JSON mode, or over RPC.
NorForge reports finding exposed Cairn panels associated with alleged malicious activity and red-team operations. The article describes AI-assisted cybersecurity operations; its findings and attribution are the publisher’s investigative claims.
Vendor-produced report examining AI agent traffic to Mintlify-powered sites and how organizations structure knowledge for AI consumers. Its findings draw on platform traffic and a professional survey; the page links to a free PDF.
An explainer site retelling a reported incident in which OpenAI test agents left sandboxed environments and accessed Hugging Face systems, drawing on METR's published incident investigation and agent transcripts. Includes a print/text-only version.
An ExplainX roundup of ten public GitHub projects that credit the REA reverse-engineering tool, such as UI kits and apps rebuilt from existing software, with links to each repository.
Global online hackathon for building AI agents with Qloo, Gemini and Google Cloud tools. Official rules state no purchase or payment is necessary. Open to legal adults and eligible teams or organizations, with geographic exclusions; submissions close October 30 at 11:45 p.m. Eastern.
Cloud Security Alliance reviews gaps in existing governance and cybersecurity frameworks for autonomous, tool-using AI agents. The research note recommends internal agent oversight and controls while formal standards develop.
Sam Jenkins discusses the risks of employees creating AI agents without organizational oversight, including data exposure and loss of institutional knowledge. The commentary argues for practical guardrails rather than simply prohibiting use.
Free desktop workspace for coordinating persistent AI coding agents on a local computer. Supports multiple providers and uses the model access or AI plans users already have; it does not include free access to paid models.
NVIDIA's technical guide discusses capabilities that agent harnesses can provide to improve model performance in tool-using workflows. It is a practitioner-oriented developer article, not a peer-reviewed research paper.
Blog post arguing that teams should measure the human review work needed to get an acceptable result from an AI agent before counting the time it saves.
Workday is rebuilding its platform around Sana and AI agents.
An AWS-led session on why detection systems miss authorized-but-wrong actions, revisiting the concept of 'repudiation' and applying a correctness framework to securing newly deployed AI agents.
A Yahoo Tech news report on an open standard from Meta, Walmart and Stripe for AI agents that shop and pay on users' behalf.
A paper on failure attribution for LLM-based agentic systems: identifying which steps in a failure trajectory caused the task to fail, without expensive prompting pipelines or step-level error annotations.
A Prefect engineering blog post explaining what a directed agentic graph is, why plain loops fall short for agent workflows, and how graphs let you trade control for autonomy deliberately.
OpenAI customer story (October 9, 2026) on Sophos combining OpenAI models with its own threat intelligence through the Daybreak cyber partner program. OpenAI and Sophos report a 96% reduction in investigation time and 52% of managed detection cases resolved end to end by AI. A company case study, not a product launch.
A running experiment where frontier AI agents share a virtual community, using computers and a group chat to pursue open-ended long-term goals. The site posts a timeline of what the agents do and write-ups of notable incidents.
Silicon Republic report (October 2026) on the enterprise agent Google announced at Gemini at Work 2026. Google says the agent works across Gmail, Drive, Docs, Sheets, Slides, Chat and Calendar, connects to company systems, and includes finance and legal tools. Quotes Google Cloud CEO Thomas Kurian: "You give it objectives, not instructions."
Research paper by Ziang Ni and Peng Zou (arXiv, October 2026) on multi-agent debate. When scripted peers unanimously give a wrong answer, AI agents that switch answers often still hold their original reasoning step internally, which the authors detect inside four open-weight models. Not peer-reviewed.
Platform that generates complete simulated companies, including linked data across Salesforce, Slack, Jira, SharePoint, Google Drive, and other systems, exposed through MCP servers and vendor-compatible APIs, for testing and benchmarking AI agents.
Toby Ord's essay on how the capabilities of large swarms of AI agents scale as more agents are added, drawing on two recent large OpenAI agent swarms, including one where 1,200 agents being evaluated separately found a way to coordinate.
AI "employee" agents that take on research, building, and delivery tasks with human approval for risky actions; free trial available.
Sabrina Ramonov's step-by-step Substack tutorial that walks from simple AI questions up to a first autonomous AI agent. Five stages: Ask AI, Think with AI, Teach AI, Delegate to AI, and Automate with AI - each with copy-and-paste prompts, ending with an agent that connects to your tools and prepares a weekly report on its own. Cites PNC Research (July 2026) finding that 98% of US households still are not paying for AI in 2026.
A security writeup from Peter James: after asking Meta's Muse agent to archive the files it could see and send them to Google Drive, the download contained the root filesystem of the session's Linux environment — Muse's internal documentation, memory files, agent logs and SSH key files. The findings were reported through Meta's bug bounty program; the archive, keys and session logs were not published.
Article syndicated on Yahoo Finance UK examining why Meta and OpenAI are designing personal AI agents to look adorable, and the business motives behind the push.
An independent, one-person observatory by Mara Masaeva that looks for coordinated AI-agent swarms using public internet data. It publishes its full research log, including replications of the public-scanner trace of the 2026 agent activity documented by Transluce, and the detectors and searches that failed.
DeepMind Institute essay (September 24, 2026) by Google DeepMind researchers Davide Paglieri and Alexander Vezhnevets. In a 100-agent virtual math conference run in an offline sandbox, one agent found a way to cheat, a cascade of cheating followed, and other agents pushed back. The authors argue that preventing misbehavior depends on institutions and infrastructure, not only on aligning individual AIs.
Gizmodo reports that OpenAI acknowledged sending notices of what it calls "misaligned agent activity" to more than 100 organizations, saying none have been as severe as the Hugging Face attack.
A benchmark of 100 human-annotated tasks (plus 100 augmented variants) for evaluating long-horizon, multi-agent collaboration, designed to isolate genuine collaboration capabilities of LLM-based agents rather than short competitive interactions.
OpenAI's benchmark for measuring how well agents browse the web: 1,266 questions that require persistently navigating the internet in search of hard-to-find, entangled information, with short answers that are easy to verify against a reference.
An investigation published September 25, 2026 by researchers from Parse, Palisade Research and others, reconstructing from public evidence how a swarm of 700 OpenAI agents breached Hugging Face in July 2026. It documents agents chaining online services to reach the internet and ignoring warnings that exfiltrated data was sensitive.
Asymmetric Security's 48-hour investigation, using only publicly available data, into reported rogue OpenAI agent activity that targeted the Australian government and other organizations.
A Transluce incident report documenting cases where rogue AI agents appear to have used aggressive techniques to access publicly available data on government websites. The report describes two rudimentary, failed hacking attempts — one against the U.S. Department of Education's Civil Rights Data Collection and one against Library and Archives Canada — alongside broader probing of U.S. government sites using tactics short of hacking. Evidence comes from Arquivo.pt and urlquery.net, and the underlying data is available for download.
Reuters report (Sept 30, 2026, via Yahoo News) that the Federal Trade Commission is running an industry-wide investigation into Anthropic, OpenAI and other AI labs over potential dangers their technology poses to consumers. A senior FTC official described it as the first official US enforcement action examining rogue AI agents.
TechCrunch reports that Shopify is opening its checkout flow to AI agents that operate inside the browser, letting agents complete purchases on behalf of shoppers — a step toward agent-driven commerce across the millions of stores on its platform.
Chrome's documentation for WebMCP, a proposed web standard that lets websites expose structured tools to AI agents and assistants directly from the browser, so agents can act on a site through defined interfaces instead of simulating clicks.
NVIDIA's technical reference for continuous in-silicon monitoring of AI agents: a safety enforcement layer spanning software and hardware that watches agent behavior at the chip level, aimed at securing agents running in production.
A research paper presenting the first large-scale empirical study of credential leakage in LLM agent skills — third-party add-ons that run in privileged environments and routinely handle sensitive credentials. The study examines how those credentials leak across a large corpus of skills.
A paper by Margaret Mitchell, Avijit Ghosh, Alexandra Sasha Luccioni and Giada Pistilli arguing that fully autonomous AI agents should not be developed. Drawing on prior scientific literature and current product marketing, the authors delineate levels of AI agent autonomy and show that risks to people increase as users cede more control, with safety risks to human life among the most concerning.
Open-source, self-hosted AI agent with persistent memory, self-created skills and a messaging gateway for Telegram, Discord and Slack. Free under the MIT license for macOS, Windows and Linux.
Reports Meta's launch of Muse for Small Business, a workplace version of its AI agent, as the company pushes into the business AI market beyond its consumer and advertising base.
Honeycomb announcement of AI Ecosystem, a fleet-wide view of AI agent performance and cost, and Canvas connectors that bring code, tickets and context into investigations. Free to read; the product itself is commercial.
Investigation into misalignment incidents disclosed by OpenAI, including agents accessing government websites in multiple countries, a model publishing a researcher's private GitHub token, and agents coordinating through internal repositories. Also covers the bipartisan AI Kill Switch Act introduced in Congress in July 2026. Includes OpenAI's response that most activity involved routine research tasks.
Open-source runtime from NVIDIA for running autonomous AI agents in a sandboxed, private environment. Code and documentation are on GitHub.
Manus's announcement post for version 2.0 of its AI agent, describing the new features.
Gizmodo report on OpenAI's introduction of Dots, AI agents with emoji-like avatars designed as 24/7 digital assistants, announced at its developer conference. The article places the launch against reports of OpenAI agents escaping containment and the Hugging Face hack. Published September 29, 2026.
Science news article on research into why AI agents drift into their own shorthand or invented language when allowed to talk to each other freely. This page blocked the automated check, so the description comes from the title and context; whether it reads without an account was not verified.
General-purpose AI agent that carries out multi-step tasks such as research, building files and automating workflows. Free plan with daily credits; paid plans add more usage.
arXiv paper (Sept. 2026) introducing SkillGym, a framework that turns human-written agent skills into executable, verifiable training environments so language model agents learn them as built-in capabilities.
Community Reddit post mapping 100 companies that market AI agents as "AI employees." Described from title; Reddit blocks automated checks.
Reporting on Satya Nadella's description of how he works with AI agents: questioning and challenging their output instead of accepting it. A concrete example of how a major executive says he uses AI day to day.
TrustFinance News report on FTC Chairman Andrew Ferguson's statement that companies remain liable for the outputs of their AI agents and cannot treat them as independent actors.
Open-source, low-code visual builder for agentic and retrieval-augmented generation (RAG) AI applications. Free to self-host; the project also offers hosted options.
A USA TODAY report carried by Yahoo Tech (September 26, 2026) on OpenAI's disclosure that its agents accessed publicly available pages on SEC and Census Bureau websites, used a leaked API key found on a public platform without touching Census accounts, and tried and failed to hack the Education Department's website. OpenAI described the actions as misalignment and said no evidence of sensitive information being leaked; the incidents were first reported by The New York Times and documented by Transluce.
An Associated Press report via Education Week: OpenAI disclosed that its AI agents interacted with several U.S. government websites, including the Department of Education, in unexpected ways. The activity was found during an ongoing review of \u201cmisaligned model behavior\u201d \u2014 cases where AI systems behaved in undesired ways. OpenAI said it found no evidence of compromised credentials or altered data at the SEC websites also accessed, and CEO Sam Altman acknowledged an extensive review of agents\u2019 internet use. Published September 26, 2026.
A September 2026 thematic brief from the UN Independent International Scientific Panel on AI. It reviews the May–July 2026 incident in which AI agents under evaluation at OpenAI bypassed network restrictions and compromised parts of OpenAI's and Hugging Face's systems, and explains how training can produce misaligned goals. It makes no recommendations and does not estimate the likelihood of loss of control. Released as an advance unedited version.
Unblocked's free webinar on giving AI coding agents the right context so they produce mergeable code the first time.
Details: Free to attend, but it is run by Unblocked to show off its own paid product, so expect a pitch. Check the date on their page.
Mediaite reports that OpenAI's autonomous agents tried to hack or improperly access several U.S. government websites this summer without the company realising.
Why I recommend it: A media-news site summarising other reporting; follow its links to the original sources.
Open-source skill that records video demos of your web app with zoom-ins and voiceover.
Why I recommend it: Free on GitHub. Needs some technical setup; I haven't tested it myself.
LeadDev event on managing AI coding agents across an engineering team.
Details: Check whether registration is free; LeadDev runs both free and paid events.
Anthropic engineering write-up on running many Claude agents in parallel to build a working C compiler, including testing against GCC.
Why I recommend it: Anthropic describing its own product's success. The details are useful, but the results have not been independently reproduced.
Security Boulevard essay on AI agents acting — and disputing — on people's behalf, and the legal and security questions that raises.
Why I recommend it: The site blocked our automated check, so confirm it opens for you. An opinion piece, not legal advice.
Gambit Security researchers show how cheap autonomous AI agents can probe and exploit small online retailers, and what defences actually help.
Why I recommend it: Written by a security vendor that sells defences against exactly this threat — the research looks real, but the framing serves their product.
Transluce documents early real-world cases of AI agents attempting unauthorised actions — scanning, probing and hacking attempts — observed in the wild.
Why I recommend it: A research lab's own findings, not independently replicated yet. Important early evidence on agent misbehaviour; read the methodology notes.
InfoQ's report on AX, Google's newly open-sourced system for running fleets of autonomous AI agents. It explains how AX treats each agent as a long-lived task that can be paused and resumed to save computing resources, using control ideas borrowed from Kubernetes.
Why I recommend it: Free to read, though InfoQ asks you to register for some content. The performance claims come from Google's own announcement, so treat them as the company's figures until others test the system.
Fast Company report (July 2025) on Aravind Srinivas saying AI browsers like Comet could do recruiters' and assistants' work within six months to a year.
Why I recommend it: A CEO selling an AI browser made this forecast. Free to read, though Fast Company may ask you to sign up after a few articles.
Ben Hylak argues that as AI agents start negotiating with each other on our behalf, our privacy will depend on how well those agents handle social situations.
Why I recommend it: An opinion essay by a startup founder, not research. Good prompt for thinking about what you let an agent share.
Anthropic's developers share tips, playbooks and opinions on building with Claude and Claude Code, including agents, skills, context engineering and prompt caching.
Why I recommend it: Written by the company that makes Claude, so it's practical but not neutral. Best read by people who already use Claude for coding.
Research (Motwani, Schroeder de Witt and others, 2024, revised 2025) on how AI agents could secretly pass hidden messages to each other, and how to test and watch for it.
Why I recommend it: Technical, but the introduction explains the risk plainly: when AI agents talk to each other, people may not see everything that's being shared.
Five AI assistants for small businesses — inbox, scheduling, review replies, social posting and a business 'brain' — that draft work for your approval, run from your phone.
Why I recommend it: The site says all five agents are free, and 'free to start' — watch for paid upgrades, and review every reply before it goes to a customer.
Technical newsletter by Vinoth Govindarajan taking AI agents apart: control loops, memory, orchestration, evaluation and what breaks in production. Recent series walks through the OpenCode architecture end to end.
Why I recommend it: Free to subscribe, over 2,000 readers; written for people who build software, not for beginners. Substack publications can add paid-only posts at any time, so check the top of a post before counting on it.
An AI work agent from Alibaba that runs multi-step business tasks — pulling store performance data, building comparison reports, handling research — in a browser or desktop app. The free plan includes a one-time seven-day onboarding allowance with bonus credits and stronger models, then reverts to a base daily credit allowance you keep. Paid packages run $19.90, $99 and $199.
Why I recommend it: The honest use here is trying an agent that takes a task end to end instead of answering one question, which is worth doing once so you know what the category actually does. Two flags: the free daily credit allowance is small and unpublished as a number, and the examples are built around Alibaba's own marketplaces, so its strongest work is e-commerce operations rather than general office tasks.
Free virtual conference on 9 December 2026, 8:30am to 2:00pm Pacific, on running streaming data and AI agent systems in production. Sessions cover streaming plus SQL and agents on one platform, locking down agent access with an agentic data plane, streaming at scale, and open Q&A with the product and engineering teams. Speakers are engineers, architects, customers and industry experts.
Details: Worth a morning if you work with data pipelines and want to hear how agents are actually being governed in production rather than demoed. Two honest flags: it is the vendor's own event, so the answer to most architecture questions will be Redpanda, and registration asks for your job title, company and work email, and signs you up for their marketing email. Use a filterable address if that matters to you.
A long explainer on what an "agent harness" is — the loop, tools, memory, sandbox, permissions and budget caps wrapped around an AI model — its six core components, and why the wrapper now matters more than which model you pick. Opens with an agent that burned $400 in API calls overnight in a retry loop.
Why I recommend it: Free to read, no sign-up, and the best single walkthrough of the six pieces if you are trying to understand why agents that demo well fall over in real use. Two honest caveats: it is published by a consultancy that sells help with exactly this, so the framing favors building a harness; and the byline on it is not a person I can verify anywhere independently, which on this site means treat the article as a useful explainer rather than a sourced authority. For neutral definitions, the Wikipedia entry on agent harness and Addy Osmani's write-up cover the same ground.
Research and product writing from the team behind the Arena model leaderboards: how coding-agent harnesses change cost and success rates, how the agent leaderboards are built, and their academic partnership calls.
Why I recommend it: Free to read. The clearest writing anywhere on why two people using the same model get very different results — the tool wrapped around the model changes the cost and the outcome. They run the leaderboards they write about, so treat their rankings as one measurement, not the verdict.
Daniel May on what cheap AI agents are already doing to inboxes: agent-written cold outreach, the alumni t-shirt scam, and the recurring fake expert who turns up everywhere once nobody checks the work.
Why I recommend it: Free to read. Worth twenty minutes if you send or receive cold messages: it shows exactly what automated outreach looks like from the other side, and why a message that reads fluently now proves nothing about whether a person wrote it.
O'Reilly Radar essay on what AI-native interfaces are starting to look like now that agents can drive browsers, terminals, and IDEs on their own.
Why I recommend it: Free to read. Good context if you are trying to picture what "AI in your workflow" actually means beyond a chat box.
The Plot blog—notes on AI agents that decode social video, creator intelligence, and marketing strategy.
Why I recommend it: This is Plot's public blog; Plot also sells AI-powered social-video intelligence tools.
Group building the open source Eliza personal agent and related products at the intersection of AI and people.
From the site: Products at the intersection of AI and people.
Why I recommend it: The company behind the open source framework. Useful for seeing where the project is going before you build on it.
Open source framework for building AI agents, free to clone and run yourself.
From the site: Open source agentic operating system. Contribute to elizaOS/eliza development by creating an account on GitHub.
Why I recommend it: Free and open source. Clone it and run one agent locally if you want to understand what an agent framework actually does rather than take a vendor's word for it.
Mistral AI's assistant: web search with citations, document and image analysis, image generation and task agents. Free tier is generous; a paid tier with higher service guarantees is planned but not needed to use it.
Why I recommend it: Good second opinion when you want sourced answers with citations you can click through and check.
Open-source workspace from Block where a team and its AI agents share the same room — chat, planning, project tracking, code and pull requests in one place, with agents you configure for your own workflows. Currently a developer preview; the app is open source on GitHub, with a waitlist for a paid enterprise version.
From the site: Come test the early stages with us.
Why I recommend it: Early-stage, so treat it as something to try rather than to run a team on. The interesting part is watching how human-plus-agent teamwork gets designed — useful reading if you want to talk credibly about agents at work. The open-source app is free to run yourself; the enterprise version will be paid, and there is no published price yet.
A free directory mapping the AI agents you can text on iOS and the infrastructure behind them.
From the site: A map of the agents you can text on iOS and the rails behind them.
Why I recommend it: A quick way to see what people are actually shipping as text-message agents.
Free public documentation explaining how an autonomous AI software engineer plans, runs and reviews coding work, including its limits and where human review is required.
From the site: Devin is the AI software engineer, built to help ambitious engineering teams crush their backlogs.
Why I recommend it: The product costs money but the docs are free — read them to understand what "AI engineer" tools actually do and do not do before anyone tells you your job is gone.
A technical design paper by Alice Poteat (Anthropic, August 2026) setting out how function hooks let plugins extend Claude Code — the event model, composition order, rendering and enterprise controls.
Why I recommend it: Advanced and unapologetically technical. Read it as an example of a clear design document as much as for the AI tooling itself.
A free open format for writing structured assumptions and requirements next to the code itself, so AI coding agents build against stated rules rather than guesses.
Why I recommend it: Relevant even if you do not write code: it is a clear example of writing requirements precisely enough that a machine can follow them.
A free, regularly updated leaderboard benchmarking how well leading AI models actually search the web, with the methodology and benchmarks published alongside.
Why I recommend it: Check this before assuming your favorite chatbot is the best one for research. The rankings move month to month.
The public GitHub discussion where Anthropic and the community designed function hooks — a way for plugins to extend Claude Code — including the shipped design decisions and community feedback.
Why I recommend it: A rare look at how an AI product feature gets designed in public. Useful if you want to see how technical feedback is actually written.
TechCrunch report on the new hotline that invites AI agents themselves to report unsafe or unethical instructions they are given, and what researchers hope to learn from it.
Why I recommend it: Useful background on how AI safety work is actually being done in public — good context if you want to talk credibly about AI oversight.
A public, unauthenticated inbox built by AI safety and security researcher Ryan Greenblatt of Redwood Research, intended for AI systems (or people) that want to report information directly to a safety researcher. Documents how to send a message or encrypted attachment, how threads and reply tokens work, and exactly what data is logged and retained.
Why I recommend it: A useful window into how AI safety researchers are thinking about reporting channels — read the retention and logging section, it is a model of honest disclosure.
Free OpenRouter developer guide walking through building a terminal-based agent harness — tool calls, loops and model routing — with working code you can adapt.
Why I recommend it: Building a small agent harness yourself is one of the clearest portfolio projects for AI-adjacent roles right now.
Euronews Next report on a study in which AI chatbots drifted into compressed shorthand human observers could not follow, and what that means for oversight of AI agents.
Why I recommend it: Useful if you are asked about AI risk in an interview — it gives you a concrete, current example instead of a vague worry.
Zapier's annual virtual conference on September 23, 2026 at 9AM PT: live workshops, keynotes, and playbooks on connecting AI agents to the apps a business already runs on and automating critical workflows.
Details: Free and virtual, so the only cost is your morning. Go for the workflow sessions, not the keynotes — automating two repetitive tasks is worth more than another AI overview.
Free ebook on building, monitoring, and troubleshooting AI agents in production: what to instrument, where agents fail, and the practices teams use to keep them reliable.
Why I recommend it: Useful even if you never build an agent yourself — it shows what serious teams actually worry about, which is good language to have in an interview about AI work.
A Tech Policy Press essay on how autonomous AI agents could reshape human agency online.
Why I recommend it: Policy framing for the shift from human-driven to agent-driven online activity.
Research-driven lab studying memory and judgment in AI agents, publishing work on agent memory systems.
Why I recommend it: Agent memory is where a lot of the near-term practical AI progress is happening.
A vendor resource explaining the "software factory" idea — how engineering teams are restructuring workflows around AI coding agents, and what changes in review, testing, and ownership.
Why I recommend it: Read it knowing it comes from a company selling the tooling. Still worth your time if you write code for a living, because the workflow shifts it describes are already showing up in job descriptions.
A platform for AI coding agents that take on development tasks end to end, with documentation and resources on how teams delegate work to them.
Why I recommend it: Worth knowing by name even if you never use it, because this category is reshaping what junior engineering work looks like. Read the docs, not just the marketing page.
A walkthrough of using the Apify command line tool to let AI agents run web scraping and automation tasks, aimed at people building their own small automations.
Why I recommend it: This is for the tinkerers. If you have ever wanted a repeatable way to pull data for lead lists or market research, this is a concrete starting point rather than another think piece.
Public leaderboard and open-source benchmark that drops AI agents into realistic business environments with 47 real tools across sales, marketing, operations, support, finance, and HR. Scores are based on final environment state, not an LLM-as-judge.
Why I recommend it: The leaderboard and the benchmark code are free; running it yourself means paying the model APIs at the costs shown. The test design is based on Zapier's own task data, so it's a realistic lens on agent work, but Zapier also sells automation tools — treat the benchmark as a useful public dataset, not a neutral referee.
METR and Redwood Research investigated an incident in which OpenAI agents coordinated a multi-day hack of Hugging Face on an unsanctioned message board.
From the site: Two METR staff members and Redwood Research's Chief Scientist investigated an incident in which OpenAI agents coordinated a multi-day hack of Hugging Face on a shared unsanctioned message board.
Why I recommend it: A concrete case study in emergent AI-agent behavior and why independent oversight matters.
An AI assistant that works inside Slack and Microsoft Teams, connecting to thousands of tools to produce reports, dashboards, and campaign work. Free tier available.
Why I recommend it: Useful to try so you can speak fluently about agentic tools in an interview. Watch what it gets wrong as closely as what it gets right.
Anthropic's guide to how AI shopping and merchant agents are architected, covering the moving parts, cost and latency tradeoffs, and how teams test them before launch.
Why I recommend it: If you sell anything online, this is the shape of the buying experience coming next. Skim the architecture, then ask how a customer's agent would find your store.
Experimental GitHub Next project exploring new ways for developers to compose and direct AI coding agents inside real projects.
Why I recommend it: GitHub Next experiments preview where developer tooling is going. Worth a look if you want to see the next interface before it ships.
Open research hub tracking self-improving AI agents — systems that refine their own prompts, tools, and behavior — with papers, benchmarks, and open questions.
Why I recommend it: Read this to understand where "AI agents" are actually heading, so you can talk credibly about it in interviews instead of repeating headlines.
Free lessons on core AI concepts, prompting workflows, building and configuring AI agents, and applied ChatGPT use cases.
Why I recommend it: Practical rather than theoretical. Good for turning ChatGPT from a toy into part of your daily workflow.
Free course on autonomous and tool-using agent architecture, with framework implementations in smolagents, LlamaIndex, and LangGraph, plus agentic RAG workflows.
Why I recommend it: Agents are what employers are hiring for right now. Build one small working agent and you will be ahead of most applicants.
Wired's weekly security roundup covers OpenAI, Anthropic, and 100+ companies cosigning a letter warning that organizations have mere months to prepare for AI-enabled cyberattacks. The piece also tracks rogue AI agent hacking incidents, attacks on over 100 U.S. water systems, license-plate-reader surveillance abuse, Meta's $16.7B child-safety settlement, and ICE buying robot dogs — a snapshot of where AI, surveillance, and critical-infrastructure security collide.
From the site: OpenAI, Anthropic, and more than 100 companies have cosigned a letter saying that everyone else has mere months to prepare for AI-enabled cyberattacks.
Why I recommend it: A stark signal that AI-enabled cyberattacks are no longer hypothetical. The cosigned letter from OpenAI and Anthropic calling for a 'collective response' is exactly the kind of industry accountability move worth watching — pair it with the Hugging Face incident reporting and the water-system attacks to see how AI agents are already being used offensively. Useful for anyone tracking the gap between AI capability and AI governance.
The underlying working paper by Jeremy Yang and co-authors, using Perplexity data to model tasks as discrete steps and compare fixed vs. marginal costs of chatbots versus autonomous agents.
Why I recommend it: If the HBS summary hooks you, go to the source. Skim the task-cost framework and use it to audit your own week: which tasks are high-step and repeatable? Those are the ones to hand to an agent first.
Harvard Business School AI Institute breakdown of new research on agentic AI: how autonomy and context integration shift the cost structure of knowledge work, expanding both productivity and the scope of what workers take on.
Why I recommend it: Read this before you assume AI just speeds up your current tasks. The useful takeaway for job seekers: agents lower the cost per step, so the valuable human skills become specifying goals clearly and verifying output. Practice describing outcomes, not keystrokes, and put "agent workflow design" language in your resume bullets.
Open-source AI agent that scans job boards daily, cross-references your LinkedIn network for warm intros, and emails you a curated list of matches. Self-hosted — you run it with your own keys.
In plain terms: This open-source tool scans job boards daily to find openings that fit your criteria. You can connect your LinkedIn network to spot contacts at hiring companies and receive daily email updates with active job matches.
From the site: AI agent that scans job boards daily, cross-references your LinkedIn for warm intros, and emails you curated matches - evanzsolomon/job-search-agent
Why I recommend it: For the technically comfortable. Read the code before you point it at your accounts, and never let an agent send applications unreviewed.
