This is Week 16 of the AI Ops Playbook — the eighth and final week of Round 2 — and it is a trending-AI week. Round 1 closed its capability gaps with image generation, voice memos, and data analysis. Round 2 closes with the three capabilities that have matured the most since: vision that can actually read your paperwork, agents that can actually run your research, and voice pipelines that can actually mine your calls. The pattern of every trending-AI week holds: the capability is not the product. The capability wired into a job you already do every week — that is the product. Here is the through-line for this drop. For two years, using AI in a small business has mostly meant typing at it. You wrote the input. This week's three templates are about the inputs you were never going to type: the shoebox of receipts nobody transcribes, the twenty open browser tabs of research nobody finishes, the forty-five minutes of sales conversation nobody writes up. Vision, agents, and voice each unlock a class of work that was stuck — not because AI couldn't help, but because getting the input into the AI was the whole job. That barrier just fell, three different ways. ## What's New This Week **Receipt & Document Data Extraction (Vision)** (Internal Ops, Beginner) — The shoebox template. Photograph a batch of receipts, attach them to one prompt, and get back a clean table plus a copy-paste CSV: vendor, date, line items, tax, tip, total, payment method — every receipt assigned to one of your real accounting categories. It is built conservatively for bookkeeping: it never guesses, writing UNREADABLE instead of a plausible wrong number; it validates that the math on the receipt actually adds up rather than silently fudging it; and it ends with a "review before entry" list so the human look goes exactly where it is needed and nowhere else. **Multi-Step Research Agent Prompt** (Strategy & Research, Advanced) — The deep-research modes in ChatGPT and Claude will happily browse forty sources for you. What they produce, unbriefed, is a book report. This template is the operating contract that turns them into an analyst: it forces a research plan before the first search, prefers primary sources, requires at least one search framed to prove each emerging conclusion wrong, tags every finding HIGH, MEDIUM, or LOW confidence, and delivers a decision brief that opens with the bottom line and argues honestly against itself. Anchored to the one thing that separates research from browsing: the decision at stake. **Sales Call Insights Extractor (Voice)** (Marketing & Sales, Intermediate) — Your meeting tools already record and transcribe every sales call; almost nobody mines them. Paste a transcript and get the full intelligence layer: every objection quoted and tagged resolved or unresolved, buying signals ranked, competitor mentions with positioning implications, deal risks with mitigations, a five-line CRM note, the follow-up email already drafted, and a coaching corner pointed at specific lines of the transcript. Its discipline is the quote-then-interpret rule — and anything inferred rather than heard is labeled INFERRED, because a sales team acts on this. ## Why These Three Together — The Input Barrier Falls Put the three side by side and they are the same template three times: work that generated no leverage because the raw material was trapped in the wrong medium. Paper. The open web. Audio. Each template is a bridge from a medium AI could not reach a year ago into the structured output your business runs on — a CSV for your books, a brief for your decision, a CRM note for your pipeline. And all three carry the same safety rail, because new capabilities fail in new ways. The vision template would rather say UNREADABLE than invent a total. The research agent tags its confidence and hunts for disconfirming evidence. The call extractor labels its inferences and refuses to reconstruct dialogue the transcript does not contain. A capability you cannot trust is a demo; a capability with its failure modes fenced is a tool. That fencing is most of what these templates are. ## Round 2 Is a Wrap — 30 More Templates, 110 in the Library This drop closes Round 2. Since late May: the retention loop, the sales motion, the people-management documents, the operating cadence, a second automation wave, four more industries, and now the new capabilities — 30 templates in eight drops, taking the library from 55 to roughly 110. Round 1 built the foundations; Round 2 built the second story. A note on cadence, one last time: this is the fifth consecutive Wednesday drop, the longest steady run since the playbook started. The streak survived because the system is boring in the best way — backlog, template, blog, newsletter, ship. That is, not coincidentally, the same lesson every template in the library teaches. What comes next: Round 3 is being shaped right now, and the strongest signal so far is industry requests — healthcare, legal, fitness, and hospitality landed in Round 2 because members asked. If you want your industry (or your most painful workflow) on the Round 3 list, reply to this week's Quick Win email and say so. The backlog is built from exactly those replies. ## Get Started All three templates are available now for Pro members: - **Receipt & Document Data Extraction (Vision)** — [open the template →](/templates/receipt-document-data-extraction) - **Multi-Step Research Agent Prompt** — [open the template →](/templates/multi-step-research-agent-prompt) - **Sales Call Insights Extractor (Voice)** — [open the template →](/templates/sales-call-insights-extractor) Start with the receipts. It is the beginner template of the three, it works with the phone already in your pocket, and it retires a chore you have been putting off since tax season. Then, next time a real decision is sitting on your desk waiting for research nobody has time to do — you know exactly which template to open.