An automated check compares the sales and installation calendars with drive times, posts problems to Slack with a suggested fix, then reruns after the office team corrects them.

ResultIn one week it caught 6 sales appointments whose drive times didn't work and suggested a revised schedule, plus an installation stop missing from the calendar and a route that couldn't fit its time slot. The owner hasn't measured time or money saved yet (owner's own report).

Every week Claude reads the site's Search Console, GA4 and Bing data and picks the pages ranking 5 to 15 with 100+ impressions; a daily Claude Code job writes, translates and publishes new posts.

ResultAverage Google position went from about 51-59 in mid-August to 6.6 on 28 September, roughly page 6 to page 1 in six weeks. The author built one of the tools used and shared no traffic or sales numbers (builder's own figures).

Every morning an AI agent checks the company's GoHighLevel CRM for uncontacted leads, overdue follow-ups and unlogged appointments, and posts them to Slack for the office team.

ResultLeads not answered within the first hour fell from 9 to 1 in a week; one report found a lead left 19 days without a reply. The owner says not all of the gain is down to AI and he has no exact ROI yet (owner's own figures).

Decoded made a 76-second documentary about a tiny snail with no camera or crew. Claude wrote the story, Gemini (Nano Banana) drew the snail and six scenes, Veo 3.1 in Google Flow animated them, and Gemini TTS read the narration.

ResultA finished 76-second short film using only free tools, with the full workflow and prompts shared.

Cybersecurity firm Sophos built AI agents with OpenAI that gather the evidence for each threat case, plan an investigation, run it and write a summary with recommended actions. Analysts review the result.

ResultAverage response time on agent-handled cases fell from about 38 minutes to 89 seconds, and 52% of cases are resolved end to end by AI.

Because security reviews were slowing projects, AGL built an agent that applies threat-modelling frameworks to generate security review documentation that architects then refine.

ResultArchitects start from an AI draft instead of a blank page; deployed in the security architecture team.

Candy's Cupcakes, a 21-person Manchester bakery that ships 100 to 350 orders a day, lost the staff member who printed every order sheet, gift tag and label by hand. The owners spent a few months experimenting with a £20-a-month Claude subscription, and now print a whole day's orders in seconds and get cupcake-topper templates ready to print quickly.

ResultA one-button order and label printing tool the owners believe saves 86 hours a month, and about £20,000 a year in staff costs.

When Maria's Pasta's part-time administrator retired, director Paul Italiano set up a Claude-based agent called Assunta to handle admin like wages and supplier invoices. He approves postings before they reach the accounts system.

ResultAbout 70% of a four-day-a-week admin role is now handled by the agent, saving more than 20 hours a week.

Founder Natia Kourdadze asked Claude Opus 5.5 for a 15-second 'showreel-style' motion graphics video about her app Diptell. Claude wrote the animation as code and rendered it. The same prompt went viral on X in late September, with many people sharing their own results.

ResultA finished 15-second launch video from one prompt, with no video editor. Others report runs from 10 to 90 minutes.

When a landline or internet service is down, Telstra's agent checks if the customer qualifies, works out the refund, watches for the fix, then pays it out. That takes the job off 1,000+ call-centre staff.

ResultRefunds are worked out and paid automatically once service is back, which takes admin off 1,000+ staff.

Instead of doing access audits by hand every so often, Telstra has an agent watching user permissions and settings in its Salesforce system all the time.

ResultConstant automatic checks replace occasional manual reviews.

Alex's team makes over 100 AI ads a month. One of their first song ads, made with Suno for the music, an AI video model and Nano Banana Pro for images, returned 5x what they spent and made more than $11K. They published a full case study of why it worked.

Result5x return on ad spend and more than $11K in revenue from one AI song ad.

Clarity Counts, a Norfolk accountancy practice with four offices, moved its 165 Direct Debit mandates to Adfin in August 2025. Adfin works alongside Xero and its AI learns when each client usually pays, then sends reminders at those times, including evenings and weekends. Debtor reviews now cover a handful of exceptions instead of a long list.

ResultInvoices went from 41 days late on average (Oct 2025) to 3-4 days late by July 2026, and Direct Debit collections grew 5x.

An agent connected to Beehiiv learns her voice from past issues and turns each new YouTube video into a draft. Nothing auto-publishes.

ResultAbout 1-1.5 hours saved per video (their own figures)

Synapxe's AgentSea lets public healthcare staff create and share agents in plain language for record summaries, procurement reviews, roster generation and productivity tasks, with governance and audit logs.

ResultEvery public healthcare worker in Singapore can now use it.

Sam Leon runs The Millennial CPA, a one-person tax practice in Richmond, Virginia. He keeps a Claude project for each client with redacted documents, has Claude draft the Excel workpaper schedule by schedule using his own rules, and uses a second Claude project to check the return against last year's and the source documents. He still types the numbers into the return himself.

ResultA corporate tax workpaper went from 6-7 hours to about 1 hour, and a 2-hour individual workpaper to about 15 minutes.

Westpac plugged five AWS agents into the core of how it processes loans and cards.

ResultMore than 150,000 hours saved.

Tiny Paws keeps the books for pet-care businesses. Its AI agent runs client audits, posts routine journal entries and builds loan schedules.

Result20–30 hours saved a month. A 3–4 hour monthly task now takes about 3 minutes (their own figures).

n8n runs lead collection, enrichment, content drafts and outreach; Claude researches and writes; Codex fixes the code.

ResultAbout 8 hours a week saved. About 60% of social posts are drafted by AI and checked before posting (their own figures).

Four agents handle content research, lead intake, competitor monitoring and newsletter first drafts. He kept the final writing and client replies human.

ResultArticles take about 90 minutes instead of 4 hours, and leads get a reply in under 2 hours (their own figures).

Solo firm owner Oskar Escobedo in El Paso moved 27 clients onto an AI-native ledger and automated monthly reporting.

ResultMonth-end work went from 18 days to 5, freeing about 90 hours a month (their own figures).

Suncorp is testing agents on parts of the claims process, and a few of those steps are heading into production.

ResultSeveral claim steps are moving from trial to everyday use.

An agile coach built an 'AI operating system' with Claude Code: n8n pulls prospects from Apollo, reads their LinkedIn posts, has Claude write personal emails and loads them into Instantly.

ResultFinding and emailing prospects now runs without them, using 60–90% less AI processing per run.

A Claude Code skill pulls the video transcript, writes the newsletter in the creator's template, then drafts it in Beehiiv.

ResultA newsletter takes about 15 minutes instead of half a day (their own figures).

A solo consultant runs lead response, content and admin with a set of AI agents costing about $160 a month.

ResultWon 41% of leads instead of about 23%, saved about 35 hours a week, and took on 2 more clients (their own figures).

Self-hosted n8n workflows handle research, first drafts and email triage, and drop everything into Notion. (Reported by a third-party blog.)

Result10+ hours a week saved for about $5 a month. Sorting email alone saves about 3 hours (their own figures).

Each morning Claude pulls his hottest leads from Follow Up Boss and drafts follow-ups. It caught a client who'd favourited a property he'd missed.

Result7 deals closed since building it; one deal came from a follow-up Claude flagged (their own figures)

After losing sleep over blown job budgets, a tradie built Formal Tradie: real-time project budget tracking with an AI agent for construction firms.

ResultLaunched as a product; results still early.

For 30 days the owner logged every task the agent handled, how long it would have taken, and the time lost to its mistakes.

Result382 tasks done and 79.6 hours saved, after 2.3 hours lost to mistakes (their own figures).

Woolworths' Olive assistant can now plan meals. Behind the scenes, eight checker agents make sure each answer is accurate, compliant and actually finishes the job.

ResultLive for 200,000+ staff, with every answer checked before it's shown.

Telstra's staff ask an internal AI assistant instead of digging through manuals mid-call.

ResultCalls are over a minute shorter on average.

Indonesia's CIMB Niaga launched two staff-facing agents on Vertex AI to help relationship managers and contact-centre staff give advice suited to each customer's goals and life stage.

ResultTwo agents are live, for account managers and the call centre.

New Zealand's ACC runs four AI tools in its claims pipeline. Staff make every final call.

ResultAI helps with claims; humans keep the final say.

Mizuho started an Agent Factory to move from building individual AI agents to producing and operating them at scale, and plans to roll the Cowork agent out to staff.

ResultProgramme launched; staff rollout targeted.

Lewie uses n8n for content, client reports, lead routing and monitoring, Supabase as a shared database, and Claude for analysis and code.

ResultEight businesses run by one person, with AI doing the repeat work.

SoftBank built a multi-AI-agent platform on its telecom-specific generative AI and began verifying autonomous network operations.

ResultStill being tested. No results yet.

Australian mortgage platform Lendi used Amazon Bedrock to build agentic AI for the refinance journey, with multi-agent orchestration and guardrails to speed up applications.

ResultRebuilt in about 16 weeks, so loan applications move faster.

A set of 12 prompts pulls hooks, threads and daily posts out of each newsletter issue.

Result15 minutes instead of 5–10 hours, and 21–28 posts a week instead of 2–3 (their own figures).

Employment Hero, serving 350,000+ businesses, says it is embedding agentic AI into regulated HR, payroll and recruitment workflows, shifting from workflow support to autonomous execution.

ResultAnnounced plan only. No results reported yet.

The founder of a reseller pricing app built a pipeline: pick a topic, generate a brief, Claude writes the article, it auto-publishes.

Result30 blog posts and 10 web pages in one go. A post went from about 2 hours to one click (their own figures).

When someone calls about bad coverage, the agent pulls network data, alarms and past incidents for that area and tells the rep what's going on in plain English.

ResultChecking takes 80% less time.

An n8n workflow pulled company signals like funding, new hires or launches, and added 1-2 personal lines to each email.

Result900 emails, 6% replied, and 10 good meetings booked (their own figures).

Japanese IT services firm Classmethod uses Claude Code across engineering and business tasks, including Google Apps Script automations.

ResultUp to 10 times more output and 90% less time coding. One task went from 24 hours to 1 hour.

Nomura Research Institute uses Claude on Amazon Bedrock to review Japanese business documents.

ResultDocument review time cut by 50%.

Philippine telco PLDT deployed UiPath agents: KAI for knowledge retrieval, Ellie for sales proposals and ERICA for automated risk assessment.

ResultRisk reviews take 97–99% less effort.

Australian payments company Send uses Relevance AI agents to take repetitive work off the team.

Result40 hours saved every week.

EvenUp builds Claude-powered tools for personal injury lawyers to draft demands, medical summaries and more.

ResultLetters that took a month take about ten minutes, and lawyers check them 75% faster.

French business bank Qonto built Claude agents that draft invoices and handle finance admin for freelancers and small firms.

ResultSending invoices takes a third of the time. Customers were spending up to 8 hours a month on finance admin.

Norwegian agency Layer used Claude connected to Supermetrics to pull marketing data from five sources into one report.

ResultA report quoted at 10 hours now takes 20 minutes, and no made-up numbers were found.

Carvana runs Claude inside Slack so whole teams can watch it work on alerts and questions.

ResultGetting an answer takes 65% less time on one team, and alerts fell 56% on another.

HubSpot's marketing team uses Claude across web development and content work.

ResultMarketing says its web and content work is 40% more productive.

The IRC tested Claude to analyse programme data from health facilities, much of it once kept in paper notebooks.

ResultEveryone in the trial worked faster, half save 4+ hours a month, and a day of analysis now takes under 30 minutes.

Rocket Money built Rowan, an agent that watches accounts and helps people manage money, with Claude.

ResultThe team's monthly coding output on the project grew 11 times in seven months.

The agent reads every quote request email. If details are missing it replies right away asking for them; if not, it texts the mechanic that a quote is ready to prepare.

ResultOwner estimated over $10,000 a year recovered from leads that used to drop off (their own figures)

The owner of a bar with golf simulators and axe throwing used Claude to plan and Claude Code to build n8n workflows for sales reports, food costs, scheduling and events.

Result80+ automations in 28 days, replacing 15–20 hours a week of manual work (their own figures).

A Claude agent scans for businesses showing signs they need help, saves them to a sheet, reads their websites and writes personal emails.

ResultAbout 400 emails a day; 23 booked sales calls in a month (their own figures)

The owner's husband built Marin Cake Studio an order system with Claude Code; a daily job drafts review requests and reminders that she approves before they send.

ResultLive on real orders 12 days after the first commit; in the first week 8 of 9 customers who got a review request left one. A sync bug once marked paid customers unpaid and was fixed (builder's own figures).