Next Great Work

Blog

  • Chatbot or Agent? What to Use, What It Costs, and Where to Begin

    The word AI now covers two quite different kinds of help.

    A chatbot helps you think. An agent can also do work.

    That difference matters. If you only want to ask questions, improve a letter,
    or talk through an idea, a chatbot may be all you need. If you want software to
    research several sources, work with files, use a browser, update a calendar, or
    carry a project through several steps, you are looking for an agent.

    You do not need to buy the most powerful service on day one. Begin with one
    small, reversible task and learn what kind of help is actually useful to you.

    Chatbot and agent: a simple comparison

    Imagine that you are planning a three-day family reunion.

    You might ask a chatbot:

    Give me a checklist for planning a family reunion in June.

    It answers with ideas and text. You still open the websites, compare locations,
    build the guest list, update the spreadsheet, and send the messages.

    You might tell an agent:

    Compare three accessible venues within 20 miles, put the hours and estimated
    costs in a spreadsheet, and draft an invitation. Do not reserve anything or
    send any message until I approve it.

    The agent may use search, a browser, files, and connected apps to complete the
    steps. It should bring the result and any important decisions back to you.

    The important distinction is not whether the screen looks like a chat. Many
    agents are controlled through a chat window. The distinction is whether the
    software only returns an answer or can also use tools and take actions.

    Four hosted services worth knowing

    Hosted services are the simplest place for most people to begin. The company
    runs the model and the computers. You open an app or website, choose what it may
    access, and pay a subscription if you need more than the free allowance.

    Prices and limits change often. The figures below were checked on October 17,
    2026, and are U.S. prices before tax.

    Grok and Grok Bot

    Grok offers ordinary chat, live web and X search, voice, and connectors. Grok
    Bot     is the more agent-like part of the service. It can use connected tools and
    carry out work rather than only discuss it.

    • A free Grok plan is available with limits.
    • Grok Bot is listed in the SuperGrok plan at $30 per month.
    • SuperGrok Plus is $100 per month for much higher usage and other extras.

    This may suit someone who already uses Grok and wants one account for research,
    conversation, and agent work. Check what a connector can read or change before
    you approve it.

    OpenAI dots

    OpenAI’s dots are always-on agents inside ChatGPT. A dot has its own cloud
    computer and browser, can connect to apps, and can keep working on several
    projects. You can inspect its work and require approval for important actions.

    • Dots are currently rolling out to eligible ChatGPT Pro users in supported
      markets.
    • The first dot is included in Pro rather than sold separately.
    • Current Pro tiers begin at $100 per month and rise with usage.

    A regular ChatGPT Plus subscription costs $20 per month and includes useful
    chat and work tools, but it does not currently include a dot. Dots make more
    sense for someone with enough ongoing work to justify a higher monthly bill.

    Meta Muse

    Meta describes Muse as a personal AI agent. It can browse, connect to apps,
    create documents, monitor goals, make purchases, and keep working after you
    close the app. Muse gives the user a view of its computer and asks for approval
    before actions such as sending messages or buying something.

    • Muse is available free with a usage limit.
    • Meta says paid subscriptions are available after that limit, but its public
      product page does not list a fixed price. Check the upgrade screen before
      agreeing to a plan.

    The free allowance makes Muse a reasonable way to see how an action-taking
    agent feels. Its many connections also mean that permission choices deserve
    careful attention.

    Claude and Claude Cowork

    Anthropic has merged Claude Cowork and chat into one Claude experience.
    Claude can answer a quick question or take on a longer task involving reports,
    documents, spreadsheets, presentations, local files, and connected apps. Some
    tasks can continue after you close your laptop.

    • Claude has a free chat plan.
    • Agent-style task handoff and scheduling are included in Claude Pro, which
      costs $20 month-to-month or the equivalent of $17 per month when paid
      annually.
    • Claude Max begins at $100 per month for heavier use.

    Claude Pro is one of the lower-cost ways to try a polished agent for document
    and knowledge work. Availability of individual Cowork features can still vary
    by device and rollout.

    What does it really cost to start?

    Next Great Work’s newsletter and guidance are free. We do not charge you to
    learn with us. The expense comes from the AI service or large language model
    you choose to use.

    A sensible starting budget is:

    • $0: learn the basics with a free chatbot or a service such as Muse that
      includes limited agent use;
    • about $20 to $30 per month: use a mainstream paid plan such as Claude Pro
      or SuperGrok when you have a real weekly project;
    • $100 per month or more: consider an always-on or high-usage agent only
      when it regularly saves enough time or creates enough value to justify it.

    Do not confuse a subscription with unlimited work. Most services have usage
    limits, and some charge separately for extra computing or model use. Read the
    current plan page before subscribing. Start monthly, keep notes on what the
    agent actually accomplishes, and cancel if the value is not clear.

    A different route: OpenClaw or Hermes Agent

    OpenClaw and Hermes Agent are not large language models. They are agent
    systems: the working structure that connects a model to tools, files,
    messaging, memory, and schedules. You still choose the model that does the
    reasoning.

    Both projects are open-source. They can use a paid cloud model, a model running
    on your own computer, or a mixture of the two. A local model can be served by
    software such as Ollama or llama.cpp.

    Why choose this route?

    • You control where the agent runs and which tools it receives.
    • With a fully local model, private prompts and files can stay on your machine.
    • The software itself can be free, and there may be no per-message model bill.
    • You can customize the agent, connect it to messaging, and keep it operating
      on your schedule.
    • A hybrid setup can use an inexpensive local model for routine work and call a
      stronger cloud model only for difficult tasks.

    What are the disadvantages?

    • You become responsible for installation, updates, backups, permissions, and
      security.
    • Local models need enough memory and storage. Official guidance starts around
      8 GB of memory for the smallest setups, while capable tool-using models often
      need 24 to 32 GB or more.
    • A model that can chat well may still be poor at using tools safely and
      reliably.
    • Local answers may be slower or less capable than leading hosted models,
      especially on an ordinary computer.
    • “No monthly model bill” does not mean no cost: hardware, electricity, setup
      time, and maintenance still count.

    OpenClaw or Hermes can be excellent for a technically confident person, a
    family member who can maintain the system, or a business that needs more
    control. They are usually not the easiest first step for someone who simply
    wants to complete one useful project this afternoon.

    Which path should you choose?

    Choose a free chatbot if you want to learn prompting, ask questions, or
    draft and revise text.

    Choose a hosted agent if you want useful work soon and prefer the provider
    to maintain the computer, model, and app connections.

    Choose a local or self-managed agent if privacy, control, customization, or
    long-term operating cost matters enough to justify technical setup and ongoing
    care.

    For many people, the best answer is hybrid: begin with a hosted service,
    learn which tasks deserve automation, and move selected private or repetitive
    work to a local system later.

    Your safest first agent assignment

    Give the agent a job that is useful, reversible, and easy to inspect:

    Find three options for [a project you care about]. Put the important
    differences, prices, and source links in a one-page comparison. Do not create
    an account, send a message, buy anything, or share my information. Stop and
    ask me before taking any outside action.

    Watch what it does. Open its sources. Correct its assumptions. Then decide
    whether it earned a second assignment.

    Sources and current plan pages

    Product features, prices, limits, and availability can change. Check the
    linked company page before paying. Next Great Work uses supervised AI in
    research and drafting, checks sources before publication, and does not receive
    payment for recommending any product named in this guide.

  • Your Experience Is the Brief: A First AI Agent Project After 65

    The best reason to use an AI agent is not to imitate someone younger or move at
    machine speed. It is to give your hard-won experience more reach.

    An AI agent can search, sort, compare, draft, and revise through several steps.
    It can remove hours of routine work. But it does not know what your experience
    has taught you to notice. It does not carry your obligations, relationships, or
    values. The useful combination is machine legwork under human judgment.

    This guide will help you choose and complete one small project safely.

    Why this matters now

    AARP’s 2026 technology research found that use of generative AI among adults 50
    and older rose from 9% in 2023 to 30% in 2025. At the same time, 60% said today’s
    technology was not designed with them in mind. Privacy, setup, support, cost,
    and clear value remain real barriers.

    A separate global survey commissioned by EY studied 2,515 people ages 60 to 85
    across 16 countries. Only 24% described themselves as quite or very familiar
    with AI, but the report also found substantial interest in learning. The lesson
    is not that older adults lack ability. It is that the tools and teaching often
    start in the wrong place.

    We start with a result that matters to you.

    Choose a project where your experience changes the answer

    Good first projects are useful, bounded, and easy to inspect. Examples:

    • turn handwritten career notes into a five-page guide for new employees;
    • organize family photographs into a timeline and list unanswered questions;
    • compare three volunteer programs against the commitments you can make;
    • create a reading plan for a subject you have always wanted to study;
    • prepare a background brief for a neighborhood or nonprofit issue;
    • outline a family storybook that grandchildren can help finish.

    Avoid a first project that involves money transfers, confidential records,
    medical decisions, legal filings, or automatic public posting.

    Step 1: Describe the finish line

    Do not begin with “What can AI do?” Begin with “What would be useful when this
    is finished?”

    Write one sentence:

    I want a __________ that will help __________ do __________.

    Example:

    I want a short guide that will help new board members understand the five
    mistakes our organization learned to avoid.

    That sentence becomes the agent’s assignment.

    Step 2: Supply the wisdom the internet does not have

    List what you know that a search engine cannot reliably infer:

    • the audience and what they already understand;
    • the history behind the work;
    • the tradeoffs that matter;
    • the people affected by the decision;
    • the values or promises that cannot be compromised;
    • examples of success and failure you personally observed.

    This is the heart of the brief. The AI may contribute speed and breadth. You
    contribute relevance.

    Step 3: Give the agent a bounded job

    Use this starter instruction:

    Help me create [result] for [audience]. First, ask me five questions
    about my experience and the outcome I want. Then propose an outline. Do not
    invent facts or quotations. Mark any claim that needs an outside source. Stop
    after the outline and wait for my approval.

    The stopping point matters. It lets you correct the direction before the agent
    produces pages of polished work based on a bad assumption.

    Step 4: Check the work in three passes

    Truth

    Are names, dates, quotations, and factual claims correct? Open the cited source.
    If the source does not support the claim, remove or rewrite it.

    Judgment

    Does the result reflect what experience taught you, including exceptions and
    tradeoffs? A neat answer can still be a shallow answer.

    Consequences

    Who could be harmed if the result is wrong? The higher the consequence, the
    stronger the independent review should be.

    Step 5: Keep the final action human

    Review the final draft yourself. If it affects another person, ask someone who
    understands the context to review it too. You—not the AI—decide whether to send,
    publish, purchase, sign, or act.

    A real example: experience gave the project meaning

    Psychologist Martin Seligman wrote that at age 83 he used AI tools to create
    illustrated adventure stories for his young grandsons. The tools helped with
    story production and images. His knowledge of the boys, his research on human
    flourishing, and his desire to strengthen the relationship supplied the reason
    the project mattered.

    That is the pattern Next Great Work will look for: technology handling a task
    while human experience determines its purpose and quality.

    Four safety rules for every project

    1. Do not paste passwords, Social Security numbers, banking details, medical
      records, or other sensitive personal information into a general AI tool.
    2. Ask for sources and open them yourself before relying on important facts.
    3. Do not let an agent send money, sign agreements, or publish in your name
      without a separate approval step.
    4. When the subject is medical, legal, tax, or investment advice, use AI to
      prepare questions—not to replace a qualified professional.

    Your first 20-minute session

    Choose one small project. Complete the finish-line sentence. List five pieces of
    experience only you can supply. Then give the starter instruction to an AI tool
    and review only the outline.

    That is enough for day one. The goal is not automation for its own sake. The
    goal is to make your next useful contribution easier to begin.

    Sources

    Next Great Work uses supervised AI in research and drafting. Sources and
    instructions are checked before publication. This article is educational and
    is not medical, legal, tax, or investment advice.