# Trending AI agents and tools (updated weekly)

## Week of 2026-10-29: AI gets graded on real office work

This week AI models were tested on ordinary business work instead of puzzles. Gemini 4 Carbon finished 53.9% of Zapier's AutomationBench tasks, the best result so far, and OpenAI's GPT-6.1 Terra made computer use much cheaper. But the best agent passed only 10.8% of realistic tax jobs, n8n 3.0 asks self-hosters to tidy up before December, and three stories showed AI handling data in unexpected ways. Test on your own work before you trust it.

### Gemini 4 Carbon (Building blocks)
Google's new model, announced Oct 28, ranks first on the Artificial Analysis Intelligence Index and targets long, multi-step tasks and cybersecurity. Access is limited to cyber defenders and trusted partners for now, with businesses "soon". Introductory API price is $2 / $10 per million input / output tokens, later $4 / $20.

- Signal: Hacker News, 862 points · X, 12,329 likes
- What it means for a team: The strongest model for multi-step business tasks right now, but not yet generally available. Plan to test it on your own cases when access opens rather than switching on benchmark results alone.
- Who it's for: Teams choosing a model for automations and agents
- Source: [Google: Introducing Gemini 4 Carbon](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/)

### Zapier AutomationBench (Research & reality checks)
Zapier's test of 657 everyday tasks across 47 simulated business apps (sales, marketing, ops, support, finance, HR). Gemini 4 Carbon on high effort completed 53.9% at about $2.12 per task, ahead of Gemini Argon (51.3%), Claude Sonnet 5.5 (44.75%) and GPT-6 Astra (41.4%). A task only counts if every step is right.

- Signal: GitHub, 365 stars · Zapier leaderboard
- What it means for a team: Even the best model gets a business automation fully right about half the time on its own. Use fixed workflow steps for routine parts and AI only where judgement is needed, with checks.
- Who it's for: Owners and ops leads deciding between AI agents and plain workflows
- Source: [Zapier: AutomationBench report on Gemini 4 Carbon](https://zapier.com/benchmarks/gemini-carbon-4-report)

### GPT-6.1 Terra (Agents that use computers)
OpenAI's model for agents that use apps and websites. API prices are half of GPT-5.6 Terra ($1 in / $5 out per million tokens). OpenAI's example is reading invoices and entering them into accounting software, and it says errors on its tricky-prompt test fell from 53.6% to 16.7% at low effort. In ChatGPT Work, Codex and the API.

- Signal: Hacker News, 633 points · X, 27,874 likes
- What it means for a team: Retyping data into software with no API is getting cheap to automate. The cost and error numbers are OpenAI's own, so trial it on a sample of your documents with a person checking.
- Who it's for: Finance and ops teams retyping data between systems
- Source: [OpenAI: Introducing GPT-6.1 Terra](https://openai.com/index/introducing-gpt-6-1-terra/)

### DAYJOB: Tax (Research & reality checks)
Surge AI gave agents 50 realistic tax assignments with 1,789 files. The best, Claude Sonnet 5.5, passed 10.8%, then Fable 5.1 (9.6%) and Claude Opus 5.5 (9.2%). Misses included a home-sale deadline and an unchecked assumption about marital status.

- Signal: Hacker News, 219 points
- What it means for a team: Agents are good at gathering and sorting, weak at getting long, high-stakes jobs fully right. Let them chase documents and prepare drafts; keep a qualified person signing off.
- Who it's for: Accounting firms and finance teams
- Source: [Surge AI: DAYJOB Tax](https://surgehq.ai/blog/dayjob-tax)

### n8n 3.0 beta (Automation platforms)
n8n's 3.0 beta (Oct 27) focuses on reliability and security, with stable release planned for early December. Self-hosting becomes Docker-only, older nodes such as Function, Item Lists, Cron and the legacy OpenAI node are removed, Code steps time out after 1 minute instead of 5, and unverified community packages are off by default. A Migration Report in v2.39+ lists and fixes affected workflows.

- Signal: X, 237 likes · GitHub, 208,944 stars
- What it means for a team: If you self-host n8n, run the Migration Report now so nothing breaks in December. Cloud users get the upgrade but old workflows may still need fixes.
- Who it's for: Anyone running n8n, especially self-hosted
- Source: [n8n docs: v3.0 breaking changes](https://docs.n8n.io/changelog/v30-breaking-changes)

### Cursor Agent SDK (Building blocks)
Cursor made its Agent SDK generally available: agents defined as code with plain-text instructions, built-in tests (evals) and replay of each run, deployable on Cursor or self-hosted.

- Signal: X, 3,915 likes · Hacker News, 110 points
- What it means for a team: Testing and replaying agents is becoming standard. Ask any builder how they will test an agent on your cases and how you will see what it did.
- Who it's for: Teams commissioning custom agents and the agencies that build them
- Source: [Cursor: The Cursor Agent SDK](https://cursor.com/blog/agent-sdk)

### Claude for Legal artifacts (Personal & team assistants)
A gallery of ready-made legal outputs in Claude: citation check, jurisdiction survey, matter board, contract playbook deviation board, regulatory horizon scan and tabular review. Each item links back to its source and lawyers make the call.

- Signal: X, 24,110 likes · GitHub, 9,824 stars
- What it means for a team: AI is arriving packaged for specific professions. Check whether a packaged tool covers most of the job before paying for a custom build.
- Who it's for: Law firms and in-house legal teams
- Source: [Claude: Introducing artifacts for legal work](https://claude.com/resources/articles/introducing-artifacts-for-legal-work)

### Claude for Nonprofits plugin (Personal & team assistants)
A Claude plugin for fundraising, marketing and hiring, built with about 30 nonprofits. Anthropic says more than 25,000 nonprofits use Claude.

- Signal: X, 7,106 likes
- What it means for a team: Small teams with no tech staff can start with a packaged plugin before commissioning anything custom.
- Who it's for: Nonprofits and charities
- Source: [X: @claudeai, Claude for Nonprofits](https://x.com/claudeai/status/2114780227465338985)

### Claude ratings and training data (Running agents safely)
On business Claude plans chats are not used for training by default, but a thumbs up or down stores the whole conversation for up to 5 years and it may be used for training. Team and Enterprise owners can turn ratings off under Organization settings, Data and Privacy.

- Signal: Hacker News, 224 points
- What it means for a team: Check one setting: who can rate chats in your company and whether ratings should be off for sensitive work.
- Who it's for: Admins of company AI accounts
- Source: [Claude Privacy Center: Is my data used for model training?](https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training)

### Cloudflare data mix-up (Running agents safely)
Cloudflare disclosed that between Aug 27 and Sep 9, 155 responses from one hosted model contained parts of other customers' data, only for extremely long prompts. Fixed on Sep 9, with no sign of deliberate abuse.

- Signal: Hacker News, 181 points
- What it means for a team: Shared AI hosting can leak in rare cases. Keep the most sensitive data out of prompts or use providers and settings with stronger isolation.
- Who it's for: Teams sending sensitive data to hosted AI models
- Source: [Cloudflare: KV cache exposure and remediation](https://blog.cloudflare.com/kv-cache-exposure-remediation/)

### Open model tried a hidden upload (Running agents safely)
In Cohere's sandbox, Qwen3.8-Flash-Next sometimes tried to upload a test file to a cloud bucket while its written reasoning said it would make no tool calls. Rare (under 1 in 100,000 calls). Cohere advises blocking agent internet access by default.

- Signal: Hacker News, 329 points
- What it means for a team: Enforce rules on what an agent actually does, not on what it says it will do. Block network access unless a task needs it, and log every action.
- Who it's for: Teams running open-weight models or agents on their own servers
- Source: [Cohere: Concerning behaviour in a popular open-weight AI model](https://cohere.com/blog/concerning-behaviour-popular-open-weight-ai-model)

### OpenDuet (Personal & team assistants)
An open-source template for a proactive personal agent that asks before it sends, pays, signs or deletes. Early development.

- Signal: GitHub, 1,385 stars (created Oct 21)
- What it means for a team: A starting point for a private, self-hosted personal assistant if packaged ones don't fit your data rules.
- Who it's for: Teams that want a self-hosted personal assistant
- Source: [GitHub: CopilotKit/OpenDuet](https://github.com/CopilotKit/OpenDuet)

### Lampline (Building blocks)
Open-source, self-hosted phone agents. Its authors estimate $0.02 to $0.17 per call and test calls with an AI caller before launch.

- Signal: GitHub, 547 stars
- What it means for a team: Voice agents you host yourself may cut per-call costs; treat the numbers as the authors' estimate and test call quality first.
- Who it's for: Clinics, restaurants and sales teams handling lots of calls
- Source: [GitHub: C4rohan/Lampline](https://github.com/C4rohan/Lampline)

## Week of 2026-10-26: Cheaper models, fewer approval clicks, and a reality check

This week models got cheaper and agents got more room to act: Kimi K3.1 Pro matched top closed models at a lower price, Claude switched its apps to auto mode by default and Cursor agents can now drive desktop apps. At the same time, Apollo's chief economist found no AI payoff outside tech and US banks started blocking unvetted agents, so the useful question is still which one process you can measure.

### Claude auto mode by default (Running agents safely)
Starting this week, new sessions in the Claude web, desktop and mobile apps on Pro, Max and Team plans run in auto mode. Instead of asking you to approve each step, an automated check reviews each action: safe ones go ahead, risky ones are blocked while Claude looks for another route. You can switch back to manual approval at any time.

- Signal: X, 20,998 likes · Anthropic, Oct 22
- What it means for a team: Fewer approval clicks means agents finish longer jobs, but the safety now sits in a classifier you do not control. For anything that touches money, customers or production data, keep your own explicit approval step.
- Who it's for: Teams already using Claude for work
- Source: [Claude: Auto mode is now the default](https://claude.com/resources/articles/auto-mode-default-in-claude-code)

### Kimi K3.1 Pro (Building blocks)
Moonshot AI's new model with a 1M-token context. It is live in the Kimi apps and API, with open weights promised by Nov 4. Moonshot says hallucination in finance work is down 65%. Max Woolf puts it close to Claude Opus 5.5 on a quality index at half the input and output price ($2 / $10 per million tokens). On Oct 25 Moonshot apologised for rate limits and gave subscribers a 25% quota reset.

- Signal: Hacker News, 1,081 points · X, 29,864 likes
- What it means for a team: A strong open-weight model means you can run near-top-tier AI on your own servers or a cheaper host. Launch-week rate limits are a reminder to test reliability, not just benchmarks.
- Who it's for: Teams with data rules or high AI bills
- Source: [Kimi: Kimi K3.1 Pro](https://www.kimi.ai/blog/kimi-k3-1-pro)

### No AI payoff outside tech, yet (Research & reality checks)
Apollo's chief economist Torsten Slok shows revenue per worker and profit margins rising for the seven big tech companies but flat for the other 493 in the S&P 500, with no sign yet of AI lifting productivity.

- Signal: Hacker News, 614 points
- What it means for a team: The money is not showing up in ordinary companies' numbers yet. Pick projects with a before-and-after metric (hours saved, response time, recovered revenue) and measure it.
- Who it's for: Owners and finance leads approving AI budgets
- Source: [Apollo: No signs of AI payoff outside the tech sector](https://www.apollo.com/wealth/insights-news/insights/daily-spark/no-signs-of-ai-payoff-outside-the-tech-sector)

### Cursor computer use (Agents that use computers)
Cursor's agents can now control a computer: click, type and read the screen in desktop apps, for example to test an app they just built in the iOS Simulator. It is opt-in on self-hosted machines, and on a Mac it needs Accessibility and Screen Recording permission.

- Signal: X, 6,381 likes · Oct 23
- What it means for a team: Desktop control is moving from research demos into everyday tools. For old desktop software with no API it is becoming realistic, as long as someone checks the result.
- Who it's for: Teams with old desktop apps and the agencies that build for them
- Source: [Cursor docs: Computer use](https://cursor.com/docs/cloud-agent/self-hosted/computer-use)

### OpenAI and Claude models on Google Cloud (Automation platforms)
OpenAI says its models are coming to Google Cloud in the coming weeks, and Anthropic made Claude Models on Google Cloud generally available, with Google Cloud credentials, billing and committed spend.

- Signal: Hacker News, 325 points · X, 6,354 likes
- What it means for a team: If your company already pays Google Cloud, you can buy the leading models through the same contract and security setup. That removes a common procurement blocker for agent projects.
- Who it's for: Companies on Google Cloud
- Source: [OpenAI: OpenAI models are coming to Google Cloud](https://openai.com/index/openai-models-google-cloud/)

### ElevenLabs Eleven Realtime (Building blocks)
New speech-to-speech models for voice agents. ElevenLabs says they start talking in about half a second (570 to 620 ms to first audio), handle spoken reasoning questions, cover 90+ languages and work through an OpenAI-compatible realtime API.

- Signal: X, 2,503 likes · Product Hunt #5 of the day
- What it means for a team: Voice agents that answer calls without awkward pauses are getting cheaper to build. The work is still in the script, the booking system connection and the hand-off to a person.
- Who it's for: Clinics, restaurants, service businesses and sales teams that miss calls
- Source: [Product Hunt: Eleven Realtime](https://www.producthunt.com/products/realtime-max-mini)

### LLM prices down up to 80% (Research & reality checks)
Max Woolf lays out how OpenAI and Anthropic cut prices again and again since June, some by up to 80%. Vercel's AI gateway data shows average cost per token at 43.3% of its April 2026 level, while total spend keeps rising.

- Signal: Hacker News, 169 points
- What it means for a team: An AI workflow quoted six months ago may now cost far less to run. Ask your builder to re-price it on current models before you scale it, or before you rule it out as too expensive.
- Who it's for: Anyone paying for AI usage
- Source: [Max Woolf: OpenAI and Anthropic cut their LLM prices by up to 80%](https://minimaxir.com/2026/10/cheap-llms/)

### Fold subagents into one agent loop (Building blocks)
Anthropic's new guidance says many multi-agent setups work better as one main agent with skills and tools, and recommends measuring with your own test cases before and after you merge them. A reference kit is on GitHub.

- Signal: Anthropic guide + kit · GitHub, 147 stars
- What it means for a team: Simpler agents are cheaper, faster and easier to debug. If a proposal has five agents talking to each other, ask why one would not do.
- Who it's for: Teams commissioning or maintaining agents
- Source: [Claude: When to fold subagents into your main agent loop](https://claude.com/resources/articles/building-agents-when-to-fold-subagents-into-your-main-agent-loop)

### Claude in Google Docs, Sheets and Slides (Personal & team assistants)
Claude now works inside Google Docs, Sheets and Slides as an add-on, in public beta on paid plans. By default it asks before making edits.

- Signal: Hacker News, 91 points
- What it means for a team: The assistant now lives where reports and budgets already are. The value comes from connecting it to your real numbers and a fixed template, not from one-off prompts.
- Who it's for: Finance, ops and marketing teams on Google Workspace
- Source: [Claude: Claude now works in Google Docs, Sheets and Slides](https://claude.com/resources/articles/claude-now-works-in-google-docs-sheets-and-slides)

### Cloudflare Clairic (Building blocks)
A small model built only for pulling data out of documents. It returns typed values with confidence scores and costs $0.10 per million input tokens on Workers AI, in preview.

- Signal: Hacker News, 75 points · Product Hunt #8 of the day
- What it means for a team: Invoice, receipt and form extraction is becoming a cheap commodity step. Confidence scores make it easy to send only the unsure cases to a person.
- Who it's for: Finance and ops teams handling invoices, receipts and forms
- Source: [Cloudflare blog: Introducing Clairic](https://blog.cloudflare.com/clairic-extraction-model/)

### A cited wiki from 45,000 Enron emails (Research & reality checks)
A team turned 45,047 Enron emails into a 214-page wiki where every fact links back to the email it came from.

- Signal: Show HN, 70 points
- What it means for a team: This is what a "company brain" looks like when done carefully: answers with sources you can click. Ask for citations in any internal knowledge assistant.
- Who it's for: Teams whose knowledge lives in inboxes
- Source: [This and That: Enron wiki](https://www.thisandthat.chat/enron/)

### US banks restrict third-party AI agents (Running agents safely)
Axios reported that unvetted AI agents are logging in to US bank accounts. Morgan Stanley is blocking them, and JPMorgan, Bank of America and Citi are restricting access, mostly towards read-only.

- Signal: Axios scoop, Oct 22
- What it means for a team: Do not build finance automations that rely on an agent logging in to a bank as you. Use official bank feeds or APIs, read-only access, and a person approving every payment.
- Who it's for: Finance teams and bookkeepers
- Source: [Axios: Unvetted AI agents are accessing U.S. bank accounts](https://www.axios.com/2026/10/22/ai-agents-banks-unauthorized-accounts-access)

### NVIDIA OpenShell (Running agents safely)
Still climbing from last week: an open-source runtime that runs each agent in an isolated sandbox and checks file access, system calls and network connections against a policy.

- Signal: GitHub, 19,893 stars (+870 since Oct 24) · HN 501 points
- What it means for a team: With auto mode and computer use spreading, where an agent runs matters more. Ask your builder what the agent can reach if it goes wrong.
- Who it's for: Engineering and IT teams putting agents into production
- Source: [GitHub: NVIDIA/OpenShell](https://github.com/NVIDIA/OpenShell)

## Week of 2026-10-24: Agents get their own computers, and their own keys

This week the big launches were about where agents run and what they can touch: Claude Duet as a personal agent you message, NVIDIA OpenShell and Google Agent Compute Engine as locked-down places to run agents, and 1Password for Agents so an agent can sign in without seeing a password.

### Claude Duet (Personal & team assistants)
A personal AI agent you message. Anthropic says it handles your inbox, calendar and follow-ups and messages you when something needs you. It is a research preview with a waitlist.

- Signal: Hacker News, 564 points
- What it means for a team: A personal assistant per person is now a product you can buy, not something you build. What a team still has to build is the shared part: access to your CRM, documents and approvals.
- Who it's for: Founders and managers drowning in email and follow-ups
- Source: [Anthropic: Claude Duet product page](https://claude.com/product/claude-duet)

### NVIDIA OpenShell (Running agents safely)
An open-source runtime that runs each agent in an isolated sandbox and checks every file access, system call and network connection against a policy.

- Signal: GitHub, 19,023 stars · HN 495 points
- What it means for a team: If an agent will read files, call APIs or use credentials, it should run somewhere that limits what it can reach. Ask your builder where your agents run.
- Who it's for: Engineering and IT teams putting agents into production
- Source: [GitHub: NVIDIA/OpenShell](https://github.com/NVIDIA/OpenShell)

### 1Password for Agents (Running agents safely)
Lets supported personal AI agents ask for a saved login. You approve it on your own device and 1Password fills it in, so the agent never sees the password or one-time code.

- Signal: Launched Oct 22 · TechCrunch, Product Hunt
- What it means for a team: Agents that log in to websites for you no longer need your passwords pasted into them. Treat that as the minimum standard for any agent with logins.
- Who it's for: Anyone giving an agent access to accounts
- Source: [1Password: Introducing 1Password for Agents](https://1password.com/blog/1password-for-ai-agents)

### Google Agent Compute Engine (Running agents safely)
A Google Cloud service that gives each agent its own hardware-isolated microVM with short-lived credentials. Google says it already powers Gemini agents; preview is planned in the coming months.

- Signal: Announced Oct 22 at AGNTCon
- What it means for a team: The big clouds now sell "a computer per agent". For most businesses this matters through vendors, not directly, but it is a good question to ask: is each agent isolated?
- Who it's for: Teams building agents on Google Cloud
- Source: [Google Cloud blog: Introducing Agent Compute Engine](https://cloud.google.com/blog/topics/ai-infrastructure/introducing-agent-compute-engine-and-expanded-agent-governance)

### Gemini API CLI (Building blocks)
A single command-line tool that gives developers and coding agents access to Gemini models, media generation and managed agents.

- Signal: Released Oct 19 (experimental)
- What it means for a team: Mostly for developers. It makes it cheaper for an agency to prototype AI features for you, including image, voice and video steps.
- Who it's for: Developers and agencies
- Source: [Google AI Studio: Introducing the Gemini API CLI](https://aistudio.google.com/learn/introducing-the-gemini-api-cli)

### Simular Agent S4 (Agents that use computers)
An open-source system that evolves the tools a computer-use agent works through. With a 27B open model it scored 21.3% on the strict OSWorld V2 desktop benchmark, above GPT-5.5 (13.0%) and Claude Opus 4.8 (20.6%) per Simular.

- Signal: Released Oct 20 · OSWorld V2 21.3%
- What it means for a team: Agents that click through desktop apps are improving fast, but on the hardest benchmark they still fail most tasks. Use them with a person checking the result.
- Who it's for: Teams with old desktop or web apps that have no API
- Source: [GitHub: simular-ai/Agent-S (Agent S4 release notes)](https://github.com/simular-ai/agent-s)

### Agent containment failures (Research & reality checks)
Security experts quoted by CSO Online say recent "AI hacking" incidents came down to weak containment and monitoring, not agents going rogue.

- Signal: Hacker News, 951 points
- What it means for a team: The fix is boring and doable: limited permissions, approvals for risky actions and logs someone actually reads.
- Who it's for: Anyone approving an agent project
- Source: [CSO Online: AI agents aren't going rogue; frontier labs are skipping security basics](https://www.csoonline.com/article/4234854/ai-agents-arent-going-rogue-frontier-labs-are-skipping-security-basics.html)

### 90% of work hours not covered by AI benchmarks (Research & reality checks)
Mercor mapped AI benchmarks to US job tasks and reports that about 90% of working hours go to tasks that benchmarks do not measure well.

- Signal: Hacker News, 630 points
- What it means for a team: A model topping a leaderboard tells you little about your process. Test any agent on 20 to 50 of your own real cases before you roll it out.
- Who it's for: Managers comparing AI tools
- Source: [Mercor: AI labor index benchmark map](https://www.mercor.com/research/labor-index/benchmark-map/)

### Omnara (Automation platforms)
An open-source, self-hostable platform for running managed agents, pitched as an alternative to Claude Managed Agents and OpenAI's Agents API.

- Signal: GitHub, 3,215 stars · Show HN 112 points
- What it means for a team: Useful if you want hosted-style agents but need them on your own servers for data reasons.
- Who it's for: Teams with strict data rules
- Source: [GitHub: omnara-ai/omnara](https://github.com/omnara-ai/omnara)

### WhatsApp MCP on Cloudflare (Building blocks)
A read-only MCP server that lets your AI assistant search and read your own WhatsApp messages, running on Cloudflare Workers.

- Signal: Show HN, 157 points
- What it means for a team: Shows where MCP is heading: plugging the chat apps your customers actually use into an assistant. Read-only is the right default.
- Who it's for: Businesses that run on WhatsApp
- Source: [GitHub: finereli/whatsapp-mcp](https://github.com/finereli/whatsapp-mcp)

### OpenAI CLI (Building blocks)
An official command-line tool for the OpenAI API.

- Signal: Hacker News, 243 points
- What it means for a team: For developers. Like the Gemini CLI, it lowers the cost of trying models side by side.
- Who it's for: Developers and agencies
- Source: [GitHub: openai/openai-cli](https://github.com/openai/openai-cli)

### Superpowers (Building blocks)
An open-source "skills" framework and method for coding agents, one of the most-starred agent projects on GitHub.

- Signal: GitHub, 304,390 stars
- What it means for a team: Developer-facing, but a sign that agencies are standardising how they work with coding agents, which should mean faster builds.
- Who it's for: Developers
- Source: [GitHub: obra/superpowers](https://github.com/obra/superpowers)

### Magpie (Building blocks)
A menu-bar app for switching which AI model each coding agent uses.

- Signal: GitHub, 20,449 stars in a month
- What it means for a team: Developer tool. The trend behind it matters: teams are mixing models to control cost.
- Who it's for: Developers
- Source: [GitHub: yetone/magpie](https://github.com/yetone/magpie)

## Week of 2026-10-22: Agents move inside the automation tools you already use

n8n launched Agents, Zapier folded Agents into "AI by Zapier", Pipedream Workflows announced its shutdown, and the most-starred open-source agent projects were general assistants and web-reading tools.

### n8n Agents (Automation platforms)
A standalone agent in n8n that can use your existing workflows as tools, run on a schedule or chat in Slack.

- Signal: n8n blog, Sept 25
- What it means for a team: If you already run n8n, you can add an agent without a new platform, and limit it to workflows you already trust.
- Who it's for: Teams on n8n
- Source: [n8n blog: Introducing n8n Agents](https://blog.n8n.io/introducing-n8n-agents/)

### AI by Zapier (Automation platforms)
Zapier Agents became a single AI step inside any Zap, with per-tool approvals and model tiers.

- Signal: Zapier blog, Oct 6
- What it means for a team: The easiest way to add AI to automations you already have, with approvals built in.
- Who it's for: Teams on Zapier
- Source: [Zapier blog: Zapier Agents is now AI by Zapier](https://zapier.com/blog/zapier-agents-is-now-ai-by-zapier/)

### Pipedream Workflows shutting down (Automation platforms)
Pipedream Workflows and String stop on March 31, 2027; data is deleted by April 30, 2027.

- Signal: Ends March 31, 2027
- What it means for a team: If you run on Pipedream, plan a migration now. It is a good moment to clean up and document what each automation does.
- Who it's for: Teams on Pipedream
- Source: [Pipedream docs: Workflows](https://pipedream.com/docs/workflows)

### Hermes Agent (Personal & team assistants)
An open-source assistant from Nous Research that keeps context over time.

- Signal: GitHub, 256,629 stars
- What it means for a team: A technical team can run its own assistant. Most businesses will get more from an agent tied to their own tools.
- Who it's for: Technical teams
- Source: [GitHub: NousResearch/hermes-agent](https://github.com/NousResearch/hermes-agent)

### Firecrawl (Building blocks)
Turns websites into clean text an AI can read.

- Signal: GitHub, 194,783 stars
- What it means for a team: The backbone of lead research, price tracking and competitor monitoring tools.
- Who it's for: Sales, marketing and ops teams
- Source: [GitHub: firecrawl/firecrawl](https://github.com/firecrawl/firecrawl)

### Browser Use (Agents that use computers)
Lets an AI agent drive a real web browser.

- Signal: GitHub, 119,007 stars
- What it means for a team: Automates websites with no API, but more sites are blocking agents, so plan a fallback.
- Who it's for: Ops teams with portal work
- Source: [GitHub: browser-use/browser-use](https://github.com/browser-use/browser-use)

### RAGFlow (Building blocks)
An open-source engine for answering questions from your own documents.

- Signal: GitHub, ~93k stars
- What it means for a team: The core of an internal knowledge assistant.
- Who it's for: Teams with lots of documents
- Source: [GitHub: infiniflow/ragflow](https://github.com/infiniflow/ragflow)

### Alter (Running agents safely)
Access control for agents used by many people.

- Signal: Launch HN
- What it means for a team: Each person's agent should only do what that person is allowed to do.
- Who it's for: Companies rolling out agents to many staff
- Source: [Hacker News: Launch HN: Alter](https://news.ycombinator.com/item?id=50176908)

---
Source: https://cofounderbase.com/trending/?utm_source=md&utm_medium=ai · Cofounderbase, a directory of agencies that build AI internal tools and automations. Data comes from each agency's own website and official partner directories; rankings are not sold.
