Good Morning! Here's what I have for you in today's newsletter:
xAI introduces Data Bot, a Team Bot that answers data questions
DeepSeek brings Creator Mode to Harness, plugins built from plain English
Anthropic adds message threads to the Claude API for faster agent replies
Understand ChatGPT’s AI watermarks and make each draft sound like you
4 new AI tools worth trying today
AI AGENTS
xAI introduced Data Bot, a Team Bot used across the company for on-demand analysis. Product managers ask about feature usage, while marketers use it to examine campaign performance.
The data science team maintains its skills, covering core metrics, A/B tests, chart styles, dashboards, and data engineering. Charts created in chat can become Databricks or Hex dashboards.
Managers review proposed memories before the Bot shares that knowledge across the team. It can also join channels and post findings without being asked.
xAI estimates it handled about 70 hours of work in one week that would otherwise have reached data scientists. Team Bots are in public beta on Teams and Enterprise plans.
“Did that campaign work?” can be a quick question with a long wait for an answer. Data Bot aims to handle routine analysis using methods maintained by the data team. xAI’s time-saving estimate comes from its own use, but the setup is worth noting: analysts define the methods, and managers approve what the Bot remembers.
AI PLUGIN
DeepSeek introduced Creator Mode in Harness, its open-source agent app. Describe a plugin in plain English, and the agent writes the code, installs it, and checks whether it works.
Demonstrations include a timeline-based video editor, a paper reader with notes and highlights, a poster workbench, a 3D generator, and a pixel-game version of the Harness interface.
Created plugins stay saved for future sessions. Creator Mode joins the app’s existing Standard, PTC, and Minimal modes.
DeepSeek shared the prompts for seven demos, noting that some builds needed several rounds of conversation. Harness remains in preview, with core plugins and APIs subject to change.
A custom tool usually starts with learning how an app’s plugin system works. Creator Mode lets users begin with what they want the tool to do instead. DeepSeek’s examples show the range, but also the back-and-forth involved. With Harness still in preview, a useful prototype may need maintenance as the app changes.
AI TOOLS
Anthropic released message threads in public beta on the Messages API. Conversations stay on its servers between turns, so apps send only new messages and the previous response ID instead of uploading the entire history again.
Claude Code already uses threads. For requests containing 500K to 1M tokens, median time to first token fell about 35%. Median request size in long conversations dropped from roughly 800 KB to 1.8 KB.
Threads stay encrypted in memory and aren’t written to disk. They expire five minutes after the last request, at which point the app must resend the conversation.
Support covers Claude Platform, Claude Platform on AWS, Microsoft Foundry, and Google Cloud. Standard token pricing still applies.
Long conversations come with baggage: every new turn can mean sending the same history again. Threads reduce that repeated upload and help Claude start responding sooner. Developers still need to handle the five-minute expiry, and smaller requests don’t mean cheaper token rates. The main benefit is less waiting during active sessions.
HOW TO AI
OpenAI is rolling out textGrain, an invisible text watermark, to eligible ChatGPT users in the EU. Its detector is initially restricted to approved researchers and expert organizations. This guide focuses on your writing voice.
Step 1: Collect your samples
Choose three pieces you wrote yourself: a post, an email, and something longer. Aim for 200–500 words each where possible. Pick writing that sounds like you normally do, rather than something heavily edited by someone else. Remove private names, contact information, and confidential details before sharing.
Step 2: Build a voice project
Create a ChatGPT project called My Voice. Paste your samples into a project chat and ask for a voice profile under 250 words covering sentence rhythm, recurring phrases, openings, punctuation, and tone. Request short quotes supporting its observations.
Read the profile and correct anything that doesn’t fit. Then paste the approved version into the project’s Instructions. Add a rule to preserve your facts and opinions without inventing experiences. Start another project chat and check that it follows those preferences.
Step 3: Start from your notes
Write three to six rough bullets containing your point, a relevant example, and what you want the reader to understand or do. Ask for a post or email around 150 words using only that material. Compare the draft with your notes and remove unsupported additions.
Step 4: Edit and check
Read the draft aloud. Change wording you wouldn’t use and add a personal detail where it helps. Then give ChatGPT your notes and edited text. Ask it to flag claims needing verification and sentences that don’t match your profile, without rewriting.
P.S. You can access all my AI courses, trainings, prompts library, and $500+ worth of AI tool discounts here.
OpenAI introduced LifeVal, a benchmark covering everyday jobs such as finding childcare, budgeting, and navigating government paperwork. Tasks draw on two time-use studies and a survey of 3,995 people. CNN reports GPT-6 Astra scored 80%, compared with GPT-5.2’s 49.3%.
Perplexity released pplx-embed-v2-context, open-weight vision-language embedding models in 9B and 0.6B sizes. They produce a compact vector for each document or chunk, with chunks retaining whole-document context. A paper accompanies the release, written with turbopuffer.
Firecrawl launched people and company search in Alexandria for agents in ChatGPT, Claude, and Grok Bot. It covers 250M+ people and 26M+ companies, including emails and employment details, with 1,000 free monthly credits.
🔔 Exa: Released Monitors V2 for alerts on new matches across web pages, businesses, products, and people. Pricing is $2 per 1,000 runs.
🔁 Rerun: Launched an open-source agent workspace that supports ChatGPT or Claude subscriptions and other models, with unlimited free seats.
🧠 Beacon: Records sessions from Claude Code, Codex, Cursor, and 20+ other coding agents, giving the next agent context to continue unfinished work. The memory layer is open source.
🗺️ Xmind: Released Xmind 2027 with an agent that researches and develops ideas from documents and maps, plus MCP access through ChatGPT and Claude.
THAT’S IT FOR TODAY
Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!
See you tomorrow :)
- Dr. Alvaro Cintas
✓ Full archive of premium guides with ready-to-use prompts
✓ Structured AI courses (step-by-step, start-to-finish)
✓ Every upcoming premium tutorial
✓ $500+ worth of AI tools