AI Workflows  /  Tools and pricing

The Agentic AI That Finishes the Job Just Got Cheap

Claude Sonnet 5 brings near top-tier agentic ability to the plan a solo practice already pays for. Here is what changed on June 30, and a document-review workflow you can run this week.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published July 3, 2026  ·  Last updated July 3, 2026

The barrier was never capability. It was price and reliability.

Professionals have been told for two years that AI agents would run their busywork. Most who tried it quit. The agent would start a multi-step job, lose the thread halfway, and hand back something you had to redo. The models that did hold the thread were the expensive ones, and paying top-tier token prices to review a lease did not add up for a one-person practice.

That equation is what changed on June 30, 2026, not the underlying idea.

What changed on June 30

Anthropic released Claude Sonnet 5 and put it forward as its most capable middle-tier model. In the company's words, "Claude Sonnet 5 is built to be the most agentic Sonnet model yet. It can make plans, use tools like browsers and terminals, and run autonomously at a level that, just a few months ago, required larger and more expensive models."

Two facts make that matter for a small practice.

• It is the default, not an upgrade. Anthropic states that Sonnet 5 "is available across all plans" and "is the default model for Free and Pro plans," and that it also runs in Claude Code and on the Claude Platform. If you pay for Claude Pro, you already have it.
• The API price dropped for the agentic tier. Introductory pricing is $2 per million input tokens and $10 per million output tokens through August 31, 2026, rising to $3 and $15 after that. For comparison, the top-tier Opus 4.8 runs $5 and $25. You are getting close to top-tier agentic behavior at a fraction of top-tier cost.

Anthropic also reports that Sonnet 5 "shows an overall lower rate of undesirable behaviors than Sonnet 4.6, and is generally safer to use in agentic contexts." Independent trackers put its agentic-coding and computer-use scores near the Opus tier, though those figures are third-party, so treat them as directional rather than official.

What "agentic" means when you are not an engineer

Strip out the jargon. An agentic model can take a job with several steps, use tools to do each step, check its own work, and keep going until the job is done. The tools are the part that is new for non-technical users. Sonnet 5 can read and write files, browse the web, operate a computer, and reach into connected apps like email, a document drive, or a calendar through what Anthropic calls connectors. You describe the outcome. It plans the steps.

For a professional, that turns a pile of manual review into a single instruction.

A workflow you can run this week: overnight document review

Here is a concrete, reproducible workflow. It suits a solo attorney or real estate agent, but the shape generalizes.

Goal. Turn a folder of incoming contracts or leases into a reviewed shortlist with flagged issues and a client-ready summary.

One-time setup, about 30 to 60 minutes. Use Claude Pro, where Sonnet 5 is the default. Optionally connect your document drive and email through connectors so Claude can read the source files and draft replies in place. No code is required. Claude Code is available if you later want a repeatable script, but you do not need it to start.

The steps.

• Drop the documents into your connected drive folder, or attach them directly.
• Give one instruction. For each document, extract the parties, key dates, and financial terms, then flag any clause that deviates from your standard template or your checklist, along with missing signatures, mismatched dates, and non-standard indemnity or termination language. Ask for one table plus a five-bullet plain-English summary per document.
• Let it run. Sonnet 5 plans the job, reads each file, and returns the table and summaries in one pass. Ask follow-up questions on anything it flags.
• Have it draft the client email. Then you read it, correct it, and send it.

The work you used to do line by line becomes a review of the model's output. That is the leverage.

The honest cost and time picture

This is where most AI write-ups oversell, so here is the restrained version.

On cost, a roughly twenty-page contract is on the order of fifteen thousand input tokens, with a few thousand tokens of output. At introductory pricing that is a few cents per document in raw API terms, and if you are on the flat Pro subscription it is included. One caveat worth knowing: some third-party analysts report that Sonnet 5 uses a new tokenizer that can raise the real token count for the same text, so treat any single dollar figure as illustrative, not a quote.

On time, do not trust a number you did not measure. The defensible framing comes from the profession, not from a vendor. Thomson Reuters projects that AI could save professionals on the order of 240 hours a year across all AI-assisted work. The Wolters Kluwer 2026 Future Ready Lawyer survey found that 62 percent of lawyers report weekly time savings in the range of 6 to 20 percent. Corporate legal adoption is climbing fast. The FTI Consulting and Relativity General Counsel Report found in-house generative-AI use jumped to 87 percent, up from 44 percent a year earlier. Cite those as context. Measure your own workflow before you claim a result.

Guardrails that keep you out of trouble

Speed without judgment is how people get hurt. Three rules.

• Human in the loop, every time. Nothing goes to a client, a court, or a counterparty without your review. The model can misread or miss a clause.
• Mind confidentiality. Confirm how your plan handles data before you upload privileged or client-confidential documents, and follow your own professional rules on AI use.
• Disclose where required. If your jurisdiction or a client relationship requires you to disclose AI assistance, do it. Honesty here is cheap and protects you.

Where this is going

The pattern is bigger than one model. The cost of an AI that can actually complete a multi-step professional task fell below the line where a solo practice can justify it, and it landed in the default plan rather than a premium tier. The competitive gap is no longer access to the capability. It is whether you build the workflow. The professionals who spend an hour this month setting up one real workflow will spend the rest of the year running it.

Browse the full AI Workflows library

Informational analysis for working professionals, not legal or financial advice. Confirm tool capabilities, pricing, and your professional obligations before relying on any workflow.