AI Teammates Stay on the Job After You Log Off
Grok Bot keeps cloud work moving, Claude Tag gives teams a shared Slack teammate, and Hermes profiles make a tuned agent portable.
What matters today
Compare Grok Bot, Claude Tag, Claude Cowork, and portable Hermes profiles, then test one bounded job for attention removed after review.
Key points
- Grok Bot runs signed-in tasks on its own cloud computer and can continue after the user closes the laptop.
- Claude Tag is the shared Slack teammate; Cowork remains an individual browser-task workspace.
- Hermes profiles can now be exported with credentials stripped and imported or distributed across a team.
- Long research runs need a specific decision, approved sources, evidence rules, and an observable finish line.
- The best agent pilot uses one bounded job, narrow permissions, one approval checkpoint, and a score based on human attention removed.
Welcome byte
The useful test for an AI agent is whether it can carry a job across tools, follow its rules, and return only when judgment is required. Grok Bot, Claude Tag, and Hermes portable profiles approach the teammate problem: Grok gives each Bot a cloud computer, Claude Tag gives a Slack channel a shared teammate, and Hermes lets teams share a reviewed agent while keeping credentials and history private.
Claude Cowork solves a separate problem. Its Chrome session preserves private browser work across Claude surfaces; it does not create Claude Tag's team context. Choose by budget and permissions, assign one job, and measure attention returned.
Quick Hits
Grok Bot is live in early beta
xAI's cloud-based Bots can use signed-in tools, continue while the user is away, and return for approval. Read more
Grok 4.6 ships for sustained knowledge work
Grok 4.6 ties GPT-5.6 Sol Max on the displayed Artificial Analysis index and keeps standard API pricing at $2 input and $6 output per million tokens. Read more
Claude's Chrome side panel becomes Cowork
Eligible users can start authenticated browser work in a saved Cowork session and resume it from another Claude surface. Read more
GPT-5.6 Luna expands Free and Go access
OpenAI is rolling out Luna as the default for Free and Go with unlimited text chats subject to safeguards and separate tool limits. Read more
OpenAI updates GPT-5.6 Sol for Plus and Pro
OpenAI describes the updated paid model as more reliable on facts, more focused, and configurable through a thought-level slider. Read more
Top AI Updates
Three deep dives with the business case and what to do this week.
1. Grok Bot plus Grok 4.6 creates an always-on knowledge-work team
Grok Bot keeps work moving after the laptop closes. Each Bot can use a cloud computer, work across signed-in tools, and request approval when needed, while Grok 4.6 handles the longer research and analysis underneath.
Start with a reversible sequence that has named inputs and a visible finish line. Research briefings, CRM hygiene, invoice collection, and follow-up preparation work well because a reviewer can inspect the sources and result.
Why it matters: A good pilot turns repeated interruptions into one review. A failed one exposes weak permissions, sources, or exception rules.
Action items:
- Confirm access through SuperGrok Heavy, Cursor Ultra, or Cursor Teams Premium.
- Choose one reversible job and define its sources, completion test, and approval boundary.
- Record autonomous steps plus review and repair time.
2. Claude's browser agent can carry the job across surfaces
A saved Cowork session expands Claude's Chrome side panel from page help into persistent work. Claude can use existing logins, navigate, type, and fill forms while preserving task history across Claude surfaces.
That matters for vendor portals, source tabs, dashboards, and form-heavy systems. The test is whether the job resumes on desktop or web without rebuilding context or restating the stopping point.
Why it matters: Preserved evidence, work state, and next steps keep browser work continuous. Authenticated reach also raises the stakes for trusted sites and confirmation rules.
Action items:
- Start one multi-tab task on Max or Team with a ledger for evidence, blockers, and next actions.
- Resume the saved session later without restating the brief.
- Check whether the ledger and work state survived.
3. Hermes profiles turn a working agent into a team-ready package
Nous Research announced portable Hermes profiles on August 12. Use an export for a one-time transfer or a Git-backed distribution for versioned team updates. Teammates install the distribution with one command.
Hermes strips auth.json and .env from exports, but exports can still contain memories and sessions. Distributions carry personality, configuration, skills, scheduled jobs, MCP connections, and plugins while keeping each person's authentication, history, and environment separate.
Why it matters: Teams can review one agent package in Git and reuse it without sharing credentials or private history. This is distribution, not central administration or inter-agent collaboration.
Action items:
- Choose one reviewed profile with a narrow job and named owner.
- Inspect one-time exports for memories, sessions, and private state.
- Use a Git-backed distribution for updates, then verify a second installation keeps credentials and history separate.
Pro Tip
Give a long-horizon research agent an evidence contract
Use this in Grok Bot, Claude Cowork, or ChatGPT Work. Replace the brackets:
You are my research analyst. Goal: [specific decision or deliverable].
Sources I trust: [list]. Sources to ignore: [list].
Process:
1. Identify the minimum set of questions that the brief must answer.
2. Gather evidence from the approved sources only.
3. Flag any claim that rests on a single source or weak evidence.
4. Deliver a one-page brief with: key findings, confidence level, open questions, and recommended next action.
Do not stop until the brief is ready for me to make a decision.
Set the format, length, and decision. Define when login barriers or conflicting evidence require a pause.
Action items:
- Replace the goal and source lists.
- Add the owner, deadline, output limit, and approval boundary.
- Check the decisive claim and weak-evidence flags.
Productivity Gem
Measure attention removed, not agent activity
Give Grok Bot or Claude Cowork one complete multi-step job with a visible finish line. Judge it by attention returned, not tabs opened or time busy. Count unexpected interruptions, review and repair minutes, and subscription cost against the manual process. Check xAI's access details .
Measure the manual baseline first: briefing, steering, review, repair, and maintenance time.
Action items:
- Define proof of completion and actions that must remain untouched.
- Log every intervention and stop at a preset review limit.
- Compare attention returned with the manual task, then keep, narrow, revise, or retire the workflow.
Health Tip
Use Apple Watch Noise alerts to protect a conference-heavy week
Apple Watch can flag when three-minute average sound reaches a chosen threshold without recording audio. Before a conference week, enable Noise notifications and give ChatGPT the schedule plus WHO safe-listening guidance. Request a plan for distance from sound sources, quiet breaks, and hearing protection.
The alert supports awareness, not diagnosis. Keep audio and Watch history out of the prompt. Take persistent ringing or hearing difficulty to a qualified healthcare professional.
This information is for educational purposes only and is not medical advice. Please consult a qualified healthcare professional before making changes to your health routine.
Kids Tip
Run a NASA paper-airplane AI prediction lab
Build a baseline paper airplane, then ask Gemini for three design changes that each alter one variable and predict distance or accuracy. Fly every design five times and compare the evidence with the predictions.
An adult should choose a safe flight area, supervise scissors, and keep names, faces, school details, and other personal information out of the chat.
Pierre's Note
The agent race just moved from “can it answer well” to “can it finish real work while I’m offline.” Grok Bot is the clearest new entrant into the workplace agents. It's joining the other great options like Claude Tag, Jack Dorsey's Buzz, Hermes Agent and others. We are now in an era where the agents can literally do anything that you need, the limiting factor becomes which one you know how to use well and how much you want to spend as they all come with wildly different costs.
They can all do what you need, so now it's time to choose your weapon based on your projected budget. Stop being distracted by shiny objects. Choose one and learn how to use it.
— Pierre
PromptHacker.ai
About the author
Pierre has spent 25+ years building growth systems across fintech, real estate, lending, campaigns, and AI workflows, with $1.5B+ in client value delivered.
Email usThree deep dives. Four useful moves. One email worth opening.
PromptHacker turns the AI firehose into practical next steps for work, health, family, and everything time keeps trying to steal.