AI Is Now Working in the Background While You're Off the Clock
The tools that used to require babysitting just got a lot more independent — and that's actually useful for people running real businesses.
If you've tried AI before and felt like you had to hold its hand through every step, you're not imagining it. Until recently, most AI tools were reactive — you type something, they answer, end of story. You had to go back, rephrase, ask the next question, and keep the whole conversation going yourself. It worked, but it felt like hiring someone who needed constant direction every thirty seconds.
That just changed.
On June 30th, Anthropic — the company behind Claude — released Claude Sonnet 5 and launched something called Claude Cowork. Here's the part worth paying attention to: Claude Cowork runs in the cloud, meaning tasks keep going after you close your laptop. You tell it what you need done, and it works on it whether you're on a job site, at dinner with your family, or asleep.
That is not a small thing. For small business owners and contractors who already work too many hours, this is worth understanding.
What "AI Agents" Actually Means
You've probably heard the term "AI agents" floating around lately. Most explanations make it sound like science fiction. Here's the plain-English version.
An AI agent is an AI that can take a sequence of actions on its own, rather than waiting for you to feed it one prompt at a time. Instead of "ask question, get answer, ask next question," an agent can be given a goal — like "research these three competitors and write a summary" — and it will figure out the steps, execute them, and bring back the finished result.
The analogy that makes sense: regular AI is like a smart employee who needs you to walk them through every task. An AI agent is like an employee who can take an assignment, go figure it out, and come back with the finished work. You're still the one setting the direction. You're just not standing over their shoulder.
The reason this matters for small business owners is that a huge portion of your day probably looks like this: researching something, sending a follow-up email, looking up a price, putting together a quote, drafting a response to a client, summarizing a meeting. None of these things require your expertise. They just require your time. And AI agents are starting to do them without you.
That shift is what all the AI news this week has been about — OpenAI, Anthropic, and others are all pointing the same direction. The assistant era is ending. The agent era is starting.
What Claude Sonnet 5 Can Actually Do Right Now
Claude Sonnet 5 is Anthropic's latest mid-tier model, released at the end of June. "Mid-tier" in 2026 means it's genuinely capable of complex reasoning, not just answering basic questions. It was designed specifically for autonomous work — things like browsing the web on your behalf, writing and running code, managing multi-step tasks, and making decisions along the way without you.
That said, this isn't magic. You still need to give it clear direction. The difference is you give that direction once instead of fifty times.
Here's a practical example. Say you're a contractor who needs to put together a materials list and rough cost estimate for a roofing job. Before, you'd open AI, describe the job, get a rough list, go back and forth editing it, copy it into your own template, format it, and send. With the newer agent setup, you describe the job once, the agent drafts the estimate, formats it, pulls in any research it needs, and hands you something close to finished. You review it and send.
That's not a fantasy use case. That's exactly what this generation of AI is being built for.
Three Places to Use This First
Not everything benefits from an agent approach. Some tasks are better handled with a simple prompt. But there are three areas where this technology is genuinely ready to save small business owners and contractors hours every week.
Client follow-up and communication. For most small business owners, follow-up emails pile up and either go out too late or not at all. An AI can draft follow-ups based on a conversation summary you give it, pull in the right information, and have a ready-to-send message waiting. You review it, you send it. No more forgetting, no more staring at a blank screen wondering how to start.
Research. Whether you're pricing a job, evaluating a subcontractor, looking up what permits are needed, or trying to figure out what competitors in your market are charging — research takes time you don't have. AI can do this legwork while you're focused on the actual work. You come back, the research is done.
Document creation. Estimates, proposals, scopes of work, change orders, job summaries — these are the documents that keep a contracting or service business running. They also take forever when you're building them from scratch every single time. AI can generate solid first drafts that you then adjust for your specific situation, cutting the time from an hour to ten minutes.
If you're a contractor who spends regular time putting together bids, proposals, or project docs, this is exactly why I built BuildOrder.ai — for contractors who need to get estimates and project documents done fast, without a stack of templates or an office manager to handle the paperwork. If that's a pain point in your business, it's worth a look.
Prompts You Can Copy Right Now
You don't need to set up a full agent system to start getting more value out of AI today. You can get a lot done with standard Claude or ChatGPT and better prompts — right now, with a free account.
The biggest mistake people make with AI is treating it like a search engine. They type a quick question and get a generic answer. To get actually useful output, you have to give the AI context — who you are, what you do, what you're trying to accomplish.
Here's a prompt to use when you need a follow-up email after a client meeting or site visit:
Try this prompt:
"I'm a [type of contractor or business owner]. I just met with a potential client named [first name]. They're interested in [describe the job or service in one sentence]. They seemed concerned about [price / timeline / a specific issue they raised]. Write me a follow-up email that's professional but not pushy. It should remind them of what we discussed, speak directly to their concern, and give them one clear next step. Keep it under 200 words."
Fill in the brackets. You'll have a usable email in seconds — something that actually sounds like a human wrote it.
Here's a prompt for when you need to put together a rough cost estimate or bid quickly:
Try this prompt:
"I'm a [type of contractor] bidding on a job. The project is [describe it briefly — square footage, scope, materials involved]. I need a rough line-item estimate I can use as a starting point. Include labor, materials, and a contingency line. Assume the work is in [your region or state]. List any assumptions you're making so I can adjust them. Format it as a table."
You'll get a rough framework in under a minute. It's not a finished quote — you'll need to verify numbers against your actual supplier pricing and local rates — but it gets you 80% of the way there instantly. That's the part that used to take an hour.
Both prompts work in Claude, ChatGPT, and most other AI tools. The rule is simple: more context in equals more useful output. Every time you skip the context, you get something generic. Every time you include it, you get something you can actually use.
The Honest Limitations
This technology is moving fast, but it has real edges.
AI agents can and do make mistakes, especially with numbers, local regulations, and anything requiring knowledge of your specific market. If you use AI to draft a cost estimate, you still verify the numbers against your real supplier pricing. If it researches a building code, you still confirm with the local authority. The AI is fast and useful — it's not replacing your judgment on the things that require your judgment.
The other thing worth knowing: some of the more advanced agent features — especially the cloud-based "keeps working after you close your laptop" capability from Claude Cowork — are still rolling out. Depending on which plan you're on, you may not have full access yet. Standard Claude and ChatGPT are powerful as-is; the agentic features are the next layer that's coming.
None of this means you should wait. The core workflow — give context, get useful output, save time — works right now with a free account and the prompts above. The agents are the upgrade. But the basics are already worth building into how you work.
One Thing to Do Right Now
Don't try to overhaul how you work in a week. Pick one task you do every week that you hate — a follow-up email, a routine estimate, a research task, a meeting summary — and try using one of the prompts above to do it with AI this week.
Do it once. See if it saves you twenty minutes. If it does, that's your proof of concept. You don't need to read five more articles. You need one win that shows you this is real.
Open Claude or ChatGPT right now, pull up that follow-up email prompt, fill in your details, and send the email you've been putting off for three days.
That's the only thing that matters today.
Tattooed Tech covers AI in plain English for veterans, contractors, and small business owners. No hype. No jargon. Just what actually works.