This week, we start with DeepSeek making moves from model hype to product reality with an agent set for release by year end. Next, ChatGPT introduces ‘branching’, which is a lot cooler than it sounds. Let’s dive in ⬇️

In today’s newsletter ↓
🧠 DeepSeek readies AI agent to rival OpenAI
🎋 ChatGPT introduces ‘branching’ in chats
💳 Visa opens agentic payments
🦆 DuckDuckGo adds newer models to Plus
🧭 Weekly Challenge: Use AI to find your dream job listing

DeepSeek says an agent is coming before the end of the year, signaling a shift from chat to action. Instead of only generating text, the agent is designed to take multi step actions with minimal prompting, then learn from past attempts. It is a bid to compete with the most capable assistants in the market and to show that efficiency gains can translate into reliable autonomy rather than only cheap tokens.

Early descriptions point to an assistant that can break down complex goals into steps, call tools, and iterate. Think research that compiles sources, spreadsheet work that fills and checks cells, or scheduling that confirms details rather than only suggesting them. The promise is less babysitting and more delegation, where you state outcomes and the system manages the in between.

Reports describe an underlying model tuned for planning, memory, and reflection. The agent would choose what to do next, consult prior runs, and adjust strategy. That requires reliable tool use and safe guardrails, including permissions, time limits, and human approval for anything that spends money or changes data. If DeepSeek keeps its efficiency edge, it could pressure rivals to improve cost and latency while keeping accuracy stable.

Agent claims often over promise. Autonomy fails when instructions are fuzzy, websites shift, or context runs out. The hard parts are durable memory, recovery from errors, and knowing when to ask a human.

DeepSeek still needs to prove real world reliability, enterprise controls, and compliance. The upside is large if it works, since many teams want repeatable workflows that run without constant supervision. If DeepSeek delivers a dependable agent, the conversation will move from chat quality to job completion, and everyone will have to respond.

ChatGPT now lets you branch a conversation from any message, so you can try a different idea without wrecking the thread you’re already on. Hover over a message, tap More actions, and choose Branch in new chat to spin up a fresh fork that inherits context up to that point. 

The feature works on the web for all logged-in users and is designed to make side-by-side explorations—like “what if we take a totally different brief?”—feel natural rather than messy. 

Branching creates a new chat seeded with the full history up to your selected turn. From there, the fork evolves independently: you can change prompts, upload different files, or steer tone and constraints without polluting your original path. It’s handy for creative drafts, code refactors, or research plans where you want to compare multiple approaches in parallel. 

Practically, it also cuts busywork—no more copy-pasting long transcripts into a brand-new chat just to test a variant. Right now the rollout is web-only; expect mobile and desktop parity later if OpenAI follows its usual release cadence.

Long chats drift. You ask for a punchier tone; later you need the formal version back. You plan a beach trip; now you want a rainy-day museum plan instead. Branching turns those “what if we tried this” moments into a simple, side-by-side test.

There are still trade-offs. Branches can multiply quickly. Give each one a short name with a purpose and date (e.g., “Resume calm tone 2025-09-05”). Star the keeper, archive the rest. Used this way, branching feels less like chatting and more like versioning your life: compare, pick, and move on.

💳 PayPal and Venmo hand out Comet invites with a free year of Perplexity Pro

🧾 Visa opens payment rails to AI agents for true agentic commerce

🎬 Try Apple’s lightning fast video captioning model in your browser

🧠 xAI’s chief financial officer steps down amid leadership churn

🦆 DuckDuckGo adds access to newer advanced models in its subscription

🎓 Delaware launches a statewide AI certification partnership with OpenAI

Follow these steps to find the job listing that fits exactly what you’re looking for in your next role:

📋 Step 1: Paste this prompt
“Act as a job scout. Return 8 open roles that fit me. Role focus: [title or field]. Skills: [skills]. Industries: [industries]. Seniority: [junior mid senior]. Location or time zone: [region]. Remote preference: [yes or hybrid]. Salary target: [range or market]. Visa needs: [yes or no]. Output a table with title, company, location, remote, salary if listed, posting age, why it fits, and a clean apply link. Explain your search strategy in one line.”

🧰 Step 2: Pick your tool
ChatGPT, Claude, Gemini, or Perplexity. For Perplexity, set focus to the web and ask it to cite company career pages.

🔎 Step 3: Add smart filters
Try one or more:

✅ Step 4: Verify and shortlist
Open each company career page, confirm the apply button works, and save your top three. Rename bookmarks with title, company, and deadline.

That’s it for this week! Will DeepSeek make another major splash on the AI market when it rolls out its agent, and do you think ChatGPT’s branching feature is a game-changer? Hit reply and let us know your thoughts.

Zoe from Overclocked