Sharp in Five

Thursday, October 15, 2026

Share this briefWhatsAppLinkedInX

The Brief

Amp deployed new agents powered by GPT-6 Astra and Sol that execute multi-step business operations and integrate via Sign in with ChatGPT. If your department relies on manual data entry or routine project coordination, you can now assign those complete workflows to an assistant rather than prompting for individual answers.

  • Meta Muse Wonder tested. Since its introduction for agentic tasks last Tuesday, new testing reveals the model excels at autonomous execution but still struggles significantly with visual design capabilities. Source
  • McKinsey deploys AI interviews. The firm now tests entry-level candidates using 10-minute AI simulations of common client scenarios, moving generative AI directly into high-stakes performance and hiring assessments. Source
  • AI deletes TV footage. An autonomous agent accidentally erased edited material for a major show, forcing a multi-year delay and providing a costly lesson in human-in-the-loop safeguards. Source
  • States sue OpenAI. Eight US states filed lawsuits alleging the company prioritized profits over child safety, a legal challenge that could force stricter usage limits on enterprise deployments. Source

Do this: If you run autonomous agents in your production environment today, configure them to require explicit human approval before they can delete any files.

Try this at work — for managers and team leads

  1. Open ChatGPT and prompt it to generate a structured presentation deck framework for your next client pitch or internal administrative proposal.
  2. Ask the assistant to draft three personalized outreach emails based directly on the key arguments identified in that generated presentation.
  3. Review the outputs for accurate tone and specific client details before exporting the final text to your email application.
The details

Hands-on

2 picks
  1. Claude

    Build 3D games with the Claude mobile app

    Using simple language prompts in the Claude mobile app, you can create functional 3D games using the Claude Opus 5.5 Max model. This workflow demonstrates how non-technical users can prototype interactive visual experiences on their phones in a few minutes.

    Simon WillisonRead
  2. ChatGPT

    Build catering pitch decks and automated outreach with ChatGPT

    Learn how to use ChatGPT to generate professional catering presentation decks and automate client outreach emails. This workflow is designed for small business owners looking to save time on administrative sales tasks without manual drafting. You can replicate this today using standard ChatGPT features in under 30 minutes.

    OpenAI (YouTube)Watch

The AI race

5 items
  1. Amp launches agents for non-coding tasks using GPT-6 Astra and Sol

    Amp is deploying new agents powered by GPT-6 Astra, GPT-6 Sol, and GPT-Live-1 that perform complex office work beyond simple programming. This marks a shift toward AI assistants that can handle multi-step business operations and integrate via Sign in with ChatGPT.

  2. OpenAI reportedly preparing to ship new models despite flagship alignment delays

    OpenAI is racing to finalize its next flagship model, which utilizes a new looping technique to significantly boost reasoning in math and coding. While the flagship faces safety delays, the company is expected to release at least one other model by the end of the year. Knowledge workers should prepare for imminent upgrades to ChatGPT's problem-solving capabilities.

  3. Google finds no evidence of AI accelerating its own research speed

    Google DeepMind reports that while AI writes 75% of their production code, human oversight is still required for 88% of changes, showing no sudden leap in AI 'self-improvement.' This suggests that for now, AI remains a productivity tool for human experts rather than a self-running replacement.

  4. OpenAI analysis highlights rapid growth of agentic AI in European businesses

    New data shows that European enterprises are increasingly adopting AI agents that can perform multi-step tasks autonomously rather than just answering questions. This indicates a shift toward more complex automation in the workplace, suggesting managers should prepare for agents to handle more operational workflows.

  5. Eight US states sue OpenAI over ChatGPT safety and child protection

    A coalition of states is suing OpenAI, alleging the company prioritized profits over safety and misled the public about risks to younger users. These lawsuits could lead to stricter age verification and usage limits that impact how companies deploy AI tools to staff.

AI at work

5 items
  1. Collaborate in real-time with your team using ChatGPT Space

    ChatGPT Space provides a shared environment where teams can pool context, documents, and conversations to work alongside AI. It allows team members to build on each other's AI interactions, making it easier to manage group projects and shared knowledge within ChatGPT.

    OpenAI (YouTube)Read the full article
  2. Figma introduces new framework for AI agents and code layers

    Figma has announced new features that define how autonomous AI agents interact with design files and is opening a beta for code-level layers. This allows designers to better integrate AI into their creative production and bridge the gap between design mockups and actual code.

  3. McKinsey interviews job applicants using AI-simulated client scenarios

    McKinsey is introducing 10-minute AI simulations to test entry-level candidates on common workplace scenarios. This signals a shift toward using generative AI for high-stakes hiring and performance assessment in professional services.

  4. Floqer launches AI agents to automate customer research for sales teams

    Floqer is deploying AI agents that automatically research B2B sales prospects to help revenue teams prepare for meetings. Sales professionals can use these agents to skip manual LinkedIn and web searches, allowing them to focus on closing deals rather than data entry.

  5. TV production delayed years after AI agent deletes edited footage

    A technical failure involving an AI agent led to the accidental deletion of edited material for a major TV show, forcing a multi-year postponement. This serves as a critical warning for managers to implement strict human-in-the-loop safeguards when deploying agents in high-stakes production environments.

    Hacker News · AI (hnrss)Read the full article

For builders

3 items
  1. Inside Uber's internal AI infrastructure and engineering agent workflows

    Uber's engineering leadership explains how they build custom developer tools and integrate AI agents into their coding cycles. While technical, it offers a rare look at how major tech firms measure the actual productivity gains of AI-assisted engineering. Managers of software teams can use these insights to benchmark their own internal AI adoption strategies.

    The Pragmatic EngineerRead the full article
  2. Testing Meta Muse Wonder for autonomous agents and design tasks

    An analysis of Meta Muse Wonder shows strong performance for autonomous agents but limitations in design capabilities when tested through Claude Code. Professionals building custom workflows should consider it for task execution rather than visual creative work.

    Lenny's NewsletterRead the full article
  3. Epic Games releases new launcher built primarily with AI coding tools

    Epic Games used 'vibe coding'—a workflow relying heavily on AI agents to write most of the code—to rebuild its desktop launcher. While the tool is faster, the company warned of potential bugs, highlighting both the speed and the quality risks of AI-led development.

Get tomorrow's brief, curated for you.

Subscribe →