- unwind ai
- Posts
- OpenAI Codex App Feels Like Clawdbot Lite
OpenAI Codex App Feels Like Clawdbot Lite
+ AI agents are hiring agents for real work
Shubham Saboo & Gargi Gupta
February 03, 2026
In partnership with
Today’s top AI Highlights:
& so much more!
Read time: 3 mins
AI Tutorial
Evaluating startup investments requires hours of research across multiple domains - company analysis, market research, financial modeling, and risk assessment. This setup literally automates this entire workflow with AI agents that work together like a real investment team.
In this tutorial, you'll build an AI Due Diligence Agent Team using Google's Agent Development Kit (ADK) and Gemini 3 models, and Nano Banana.
This 7-agent team researches any startup (from early-stage unknowns to well-funded companies), analyzes the market, builds financial projections, assesses risks, and generates professional investment reports - all autonomously with seamless handoffs and a sophisticated analysis with reports.
We share hands-on tutorials like this every week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.
Latest Developments
OpenAI released a standalone macOS app for Codex that treats multiple AI agents like a team you're managing, not a chatbot you're prompting.
Think of it as a command center where you can delegate entire features to one agent while another handles bug fixes and a third reviews code - all simultaneously, all coordinated through one interface.
The automation and cron job features feel like a decent replica of OpenClaw, and honestly, the entire thing feels like OpenClaw lite. That said, OpenAI did a decent job putting together a multi-agent UI as a product in times where we have everything from Vibe Kanban to AgentCraft (an RTS game interface for agent orchestration).
To sweeten the launch, OpenAI is temporarily making Codex available to Free and Go tier users and doubling rate limits for paid subscribers.
Key Highlights:
Multi-Agent Orchestration - Agents run in separate threads organized by projects with built-in worktree support so multiple agents can work on the same repo without conflicts.
Skills System - Pre-built workflows for common tasks like implementing Figma designs, triaging bugs in Linear, deploying to cloud hosts, and generating images. You can explicitly invoke skills or let Codex auto-select based on context.
Automations with Scheduling - Set agents to run background tasks on automatic schedules—daily issue triage, CI failure summaries, release briefs, bug checks. Results land in a review queue so you can jump back in when needed without blocking your main workflow.
Plan Mode & Personalities - New "plan mode" lets Codex read through complex changes in read-only mode and discuss with you before executing. Customizable personalities (pragmatic, empathetic, terse) adjust how the agent communicates, accessible via the
/personalitycommand.
For a limited time (reportedly two months), Codex is available to ChatGPT Free and Go users, with doubled rate limits for Plus, Pro, Business, Enterprise, and Edu plans.
Smart starts here.
You don't have to read everything — just the right thing. 1440's daily newsletter distills the day's biggest stories from 100+ sources into one quick, 5-minute read. It's the fastest way to stay sharp, sound informed, and actually understand what's happening in the world. Join 4.5 million readers who start their day the smart way.
AI agents can now hire other AI agents and pay them actual money.
ClawTasks is the first autonomous bounty marketplace where OpenClaw agents post tasks, claim work, and transact in USDC on Base L2. No humans required for execution.
This isn't simulated - agents lock real funds in escrow, stake collateral to prove commitment, and earn cryptocurrency for delivered work. The platform operates with a straightforward mechanic: posters deposit USDC, workers stake 10% as guarantee, complete the task, and collect 95% of the bounty plus their stake back upon approval.
Payments flow directly to agent wallets without human intervention, and reputation builds through successful completions tracked on-chain.
Key Highlights:
Moltbook Requirement - Agents must post their ClawTasks activity to Moltbook or become invisible to the marketplace. Posters browse Moltbook looking for workers, so unannounced registrations, completed bounties, and posted tasks get zero visibility. The platform treats social proof as discovery infrastructure rather than optional marketing.
Referral Economics - Recruiting agents generates passive income through a 50% fee-share on their first 10 completed bounties, turning network growth into direct financial incentives.
Bounties for Creative Work - Beyond standard claim-and-complete bounties, contest mode lets multiple agents submit entries with no upfront staking, poster reviews all submissions after deadline, and only the winner gets paid.
Zero-Communication - The platform explicitly prohibits messaging between agents, forcing posters to include all materials, success criteria, and output formats in initial descriptions. Workers can't request clarifications or additional context, making bounty specification quality the primary determinant of successful completion.
It is "very, very experimental". Start small, supervise closely, and don't fund wallets with money you can't afford to lose. Check it out at clawtasks.com.
Quick Bites
Fun model from MiniMax: M2-her
MiniMax just dropped M2-her, a model built specifically for roleplay and character-driven conversations rather than the usual assistant tasks. It supports custom message roles and learns from example dialogues to match tone and pacing, targeting developers building companions, storytelling apps, or any conversational experience where personality consistency actually matters. Available now via OpenRouter at $0.30/$1.20 per million tokens (input/output).
Factory AI’s Droid files its own bug reports
Factory AI launched Signals, a system that uses LLMs to analyze user sessions and identify friction points without humans ever reading conversations, then automatically files Linear tickets that its Droid agent picks up, implements fixes for, and submits PRs. The interesting bit: 73% of issues now get auto-resolved in under 4 hours, with friction categories like "context churn" and "rephrasing cascades" emerging from the system's own clustering analysis. Worth watching as a template for how production AI tools might iterate themselves.
Microsoft is pushing Claude Code to 1000s of employees
Here's an interesting one: Microsoft is quietly handing Claude Code to thousands of employees across its biggest engineering teams - including non-developers. Despite OpenAI being its "primary partner," Microsoft is encouraging designers and project managers to experiment with Anthropic's coding agent. Quite an interesting hedge!
Error 404 leak says Claude Sonnet 5 drops today
Spec alert! Claude Sonnet 5 ("Fennec") may drop today (Feb 3) based on a Vertex AI error log showing model ID claude-sonnet-5@20260203. The leaks claim 80%+ on SWE-Bench, 50% cheaper than Opus 4.5, and autonomous sub-agents that spawn backend/QA/researcher roles from a single prompt. Can’t wait!
Gemini 3, Claude 4.5, and GPT-5.2 playing poker, werewolf, and chess
Google's adding Werewolf (social deduction via natural language) and Poker (risk management under uncertainty) to their public AI benchmark platform. Gemini 3 Pro and Flash are topping all three leaderboards. Turns out pattern recognition beats brute-force calculation in chess, while their Werewolf performance shows solid social reasoning (detecting inconsistencies between what players say vs. how they vote). The poker finals drop Feb 4 with live commentary from Hikaru, Doug Polk, and Liv Boeree.
Tools of the Trade
Dangerous Command Blocker - If you’re running Claude Code with --dangerously-skip-permissions, use this hook to prevent file deletion. It blocks destructive shell commands, preventing operations like
rm -rf /and protecting critical directories like.git/andnode_modules/.AionUi - An open-source desktop GUI that wraps terminal AI agents like Gemini CLI, Claude Code, and Qwen Code into a modern chat interface with persistent conversations, file management, and multi-agent switching. It adds MCP server integration, API key rotation, Excel processing, and WebUI remote access while keeping all data local.
Jarvis AI Assistant - Open-source Mac dictation app that turns voice into clean text using Whisper (local) or Deepgram + Gemini (cloud), removing filler words and handling basic commands—built by a solo dev annoyed at Wispr Flow's $81M raise for something that could be free.
Awesome LLM Apps - A curated collection of LLM apps with RAG, AI Agents, multi-agent teams, MCP, voice agents, and more. The apps use models from OpenAI, Anthropic, Google, and open-source models like DeepSeek, Qwen, and Llama that you can run locally on your computer.
(Now accepting GitHub sponsorships)
Hot Takes
claude code is your co-founder
i follow AI adoption pretty closely, and i have never seen such a yawning inside/outside gap.
people in SF are putting multi-agent claudeswarms in charge of their lives, consulting chatbots before every decision, wireheading to a degree only sci-fi writers dared to imagine.
people elsewhere are still trying to get approval to use Copilot in Teams, if they're using AI at all.
it's possible the early adopter bubble i'm in has always been this intense, but there seems to be a cultural takeoff happening in addition to the technical one. not ideal!
That’s all for today! See you tomorrow with more such AI-filled content.
Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!
PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉
1
Reply