• AI Valley
  • Posts
  • DeepSeek’s ‘self-improving’ AI agent

DeepSeek’s ‘self-improving’ AI agent

PLUS: OpenAI announces AI-powered hiring platform

Author

Barsee
05 Sep

Together with

Howdy. It’s Barsee again.

Happy Friday, AI family, and welcome back to AI Valley.

Today’s climb through the Valley reveals:

  • ChatGPT experiments with a max-thinking mode

  • OpenAI now lets you branch chats

  • OpenAI announces AI-powered hiring platform

  • OpenAI set to start mass production of its own AI chips

  • DeepSeek’s ‘self-improving’ AI agent

  • Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

DELVE AI

Image Credit: Devle AI

Getting compliant doesn’t have to take months. With Delve, it takes 15 hours.

SOC 2, HIPAA, ISO, and more, handled automatically with AI.

That’s why 500+ of the fastest-growing AI companies (like Lovable, Cluely, Wispr Flow) already trust Delve for compliance.

But compliance is only half the story. The other half is enterprise reviews. Delve’s team has sat through hundreds of hours of security questionnaires—every pointless checkbox, every outdated SaaS-era question, every deal delayed because of them.

Getting compliant doesn’t take months anymore. With Delve, it takes 15 hours, and today, alongside Bland, they are rethinking security questionnaires.

📖 Read the manifesto here 

⚡ Book a demo here with code AIVALLEY1KOFF for $1,000 off + white-glove onboarding.

*This is sponsored

THROUGH THE VALLEY

Image Credit: Tibor

OpenAI is testing a new “Thinking Effort” control in ChatGPT, which lets users choose how much reasoning power the model applies. Levels range from Light (5) to Max (200), with steps like Standard (18) and Extended (48). The highest tier, called “Juice 200,” is only available to Pro and Enterprise users because of its high compute cost. Other experiments include showing the active model in the composer, a collapsed tool menu, and model visibility in the Plus menu. The goal is to make ChatGPT smarter while keeping the experience simple.

Image Credit: OpenAI

OpenAI has added a new feature to ChatGPT that lets you branch a conversation into a separate thread. This means if you’re in the middle of learning something (like how recommendation algorithms work), you can branch off to ask side questions without derailing the main lesson. Once you’re done, you can return to the original thread and continue where you left off. The update makes it easier to explore related ideas, avoid losing context, and keep conversations more organized and focused.

OpenAI is creating a hiring marketplace powered by AI and introducing fluency certifications to connect workers with employers while boosting skills. The “OpenAI Jobs Platform,” planned for mid-2026, will match candidates to companies using AI, with separate tracks for small businesses and local governments. A companion project, “OpenAI Academy,” will begin pilots in late 2025 and aims to certify 10 million Americans by 2030, starting with Walmart as a partner. The move challenges LinkedIn and aligns with U.S. efforts to expand AI education and workforce training.

OpenAI will produce its own AI chip next year in partnership with U.S. semiconductor giant Broadcom, aiming to meet rising demand and reduce reliance on Nvidia. Broadcom CEO Hock Tan confirmed a $10 billion order from a new customer, identified as OpenAI, with shipments starting in 2026. The chip will be used internally rather than sold to others, following the path of Google, Amazon, and Meta. Broadcom shares have surged more than 30% this year, boosted by custom AI chip demand.

Chinese startup DeepSeek is developing a new AI agent designed to carry out multi-step tasks on its own and improve from past actions. Founder Liang Wenfeng plans to release it in Q4 this year, even as the company’s R1 successor faces delays. The system is meant to handle complex workflows with little user input, joining a wave of agent-focused tools like ChatGPT Agent and Anthropic’s Claude for Chrome. After R1’s surprise impact last year, expectations are high for another major release.

Image Credit: Forbes

Top tech executives (Zuck, Sam Altman, Tim Cook, Sergey Brin, Sundar Pichai, Bill Gates, Alexandr Wang, etc) joined Trump at the White House to announce new AI education efforts. Microsoft will give US college students a free year of Microsoft 365 with Copilot and fund teacher grants and LinkedIn Learning courses. Google pledged $150 million for AI training, Amazon promised programs for millions of learners, and Anthropic committed $1 million to K-12 cybersecurity and shared a free AI curriculum. At dinner, Mark Zuckerberg and Tim Cook cited $600 billion in U.S. investment, and Sundar Pichai pledged $250 billion.

TRENDING TOOLS

  • Nebius - The cloud that gets AI workloads. Switch to Nebius and get up to 3 months off your contract *

  • Receiptor AI - Bookkeeping on autopilot with AI

  • EmbeddingGemma - Search through your emails, notes, and web pages by asking plain-English questions

  • Ghost - A single editor that blends Cursor, Figma, and PowerPoint into one powerful workspace

  • AIvilization - Watch thousands of agents simulate future human–AI civilizations, upload your own digital self, or create custom personalities and see them interact and evolve

  • Ada - Your personal AI data analyst

(*) signifies sponsored tool

THINK PIECES / BRAIN BOOST

(*) signifies sponsored tool

THE VALLEY GEMS

What’s trending on social today:

THAT’S ALL FOR TODAY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.