AI Tools for Small Business: I Gave Jarvis One Morning and Five Real Jobs

I run a small gift-and-homeware shop. Not an online empire — a real shop with a bell on the door, a counter, and a list of jobs that never gets shorter. This morning the list had five items on it, and I gave myself one rule: every job had to go through Jarvis, the AI tools directory demo at directory.tempomailusa.com, before I was allowed to do it any other way. Not because I believe a directory can run a shop. Because I wanted to find out whether a directory built for founders, creators and researchers has anything in it for the person who has to answer the bad review before lunch.

The five jobs: answer a two-star review without sounding defensive. Write descriptions for ten new products. Draft a week's worth of social posts. Find out what the shop across the street charges for the same candles. Send the monthly newsletter. Four hours, one directory, and a notebook. Here is what the morning actually did.

The three stacks, read by someone who sells candles

Before any of the jobs, I read the three curated stacks, because the stacks are the directory's answer to the question every small-business owner actually asks: don't show me 33,379 tools, show me the three I need. The demo calls itself an index of 33,379 verified AI tools across 27 categories, updated November 8, 2026, with 49 milliseconds average latency and 1,489 open source models, all quoted as the demo's own figures. Thirty-three thousand tools is not a catalogue. It is a warehouse with the lights off. The stacks are supposed to be the torch.

The Solo Founder Stack is "Full-Stack Product Builder": "from natural language prototype to live production deployment with automated database and billing", with Cursor 0.45, v0 by Vercel and Claude 3.5 Sonnet. The Content Production Stack is a "High-Fidelity Media Studio" that will "generate broadcast-grade video, lifelike voiceovers, and marketing assets in minutes" — Runway Gen-3, ElevenLabs, Midjourney v6.1. The Research Lab Stack does "Autonomous Synthesis & Search": "synthesize peer-reviewed papers, extract structured insights, and verify factual assertions" — Perplexity Pro, DeepSeek V3, CiteTrue.

Read those three descriptions again, slowly, as a person who sells candles. One stack builds software products. One stack makes broadcast-grade video. One stack reads peer-reviewed papers. None of them answers a review, writes a product description, posts to social media, checks a competitor's price, or sends a newsletter. The stacks are not bad. They are precise, well-chosen, and clearly built for three real personas — a founder shipping an app, a creator making media, a researcher reading papers. The problem is that none of those personas is me, and the directory never pretends otherwise. It just never offers a fourth stack for the person whose morning is five small jobs and a bell on the door. That is the honest frame for everything that follows: this is not a review of whether the stacks are good. It is a test of whether they are mine.

Job one: the two-star review

The review came in yesterday evening: two stars, "lovely shop, but the candle I bought smells nothing like the description and the staff seemed rushed." Fair on the second half — Saturday was chaos — and arguable on the first. The job is to reply in a way that sounds like a human who read it, not a template that apologises in bulk.

I started with the Content Production Stack, because a reply is writing, and the stack is the directory's media answer. Runway makes cinematic video. ElevenLabs clones voices. Midjourney synthesises 8K visuals. None of them writes a seventy-word reply to an unhappy customer, and none of them claims to. The stack is a media studio, and a review reply is not media. Dead end in under a minute — but a useful one, because it tells you what the stack is for by showing you what it is not for.

So I went to the catalogue, and this is where the morning's first real observation lives. The front page shows twelve cards, all added a week ago: one Free (GitHub Pages), ten Freemium, one Paid (Universal Gcode Sender, a Java-based G-code sender for GRBL machines — a tool for running a CNC machine, which tells you how wide this index is). Ten of the twelve cards carry the exact same two lines of description: "AI-generated personalised plans" and "Recommendations from preferences". Ten. The November 7 post in this series counted five; today I counted ten, and I am flagging the change rather than explaining it, because I do not know whether the seed refreshed or the earlier count was short. Either way, ten of twelve cards on the freshest page in the directory describe themselves with the same two sentences, and those sentences describe nothing — not a job, not a price, not a limit.

Four of the twelve are gift finders. Four different tools that all do "AI finds gifts for you", on the same page, on the same morning I need a tool that answers a review. I sell gifts. The directory knows gifts. It just does not know review replies, or at least not on its front page. I found writing tools in the catalogue eventually — the index is 33,379 tools deep, and search exists — but the front page is supposed to be the directory's best foot forward, and its best foot forward is four gift finders and ten copies of one sentence. The review got answered, in the end, by me, in about six minutes, with a draft I wrote myself and a tone I checked twice. The directory contributed a search box and a lesson: the front page is organised around recency, and recency does not care what your morning needs.

Job two: ten product descriptions before lunch

Ten new products arrived on Tuesday — candles, diffusers, a run of ceramic mugs — and every one of them needs a description before it can go on the shelf or the website. This is the job I most expected the directory to solve, because product descriptions are the single most common AI writing job a small shop has.

It did not solve it, and the reason is instructive. The Content Production Stack is the closest thing the directory has to a writing answer, and it is not a writing answer at all: Runway generates video, ElevenLabs generates voice, Midjourney generates images. A product description is none of those. It is sixty words that tell a customer what a candle smells like without lying about it, and the stack built for "marketing assets in minutes" has no tool that writes sixty honest words. The stack's own description gives it away — "broadcast-grade video, lifelike voiceovers, and marketing assets" — and I respect a stack that knows what it is. I just wish the directory had a fourth stack that knew what I am.

The catalogue did better here, but only after I stopped browsing and started searching. The twelve front-page cards are the newest arrivals, not the most useful, and the useful tools for this job live deeper in the index, behind a search box and a category filter. That is the directory working as designed — 27 categories, a search box, 33,379 tools — but it is also the directory admitting that its front page is not for people with jobs. The front page is for people with curiosity. The search box is for people with a deadline.

My opinion, stated plainly, because a job test that only describes is a diary, not a verdict: the directory's biggest gap for a small business is not that it lacks writing tools. It has them — thousands of them, presumably. The gap is that nothing on the front page points at the job. The stacks point at personas, the catalogue points at recency, and the shop owner has to already know what to search for before the directory can help. A directory that only helps people who already know the answer is a phone book, not a guide. The descriptions got written — ten of them, in about forty minutes, with a writing tool I found through search and a rule I set myself: every description had to name one true thing about the product (the wax, the burn time, the weight of the mug) before it was allowed one nice adjective. The directory found the tool. The rule was mine.

Job three: a week of social posts

Seven posts, one for each day, each with a photo, a caption and a reason to exist. This is the job where I expected the Content Production Stack to finally earn its keep — it is a media studio, and social posts are media — and it half did. Midjourney can make the image. ElevenLabs can voice the reel. Runway can cut the video. What none of them does is decide what to post on Thursday, or write the caption, or know that Tuesday is the day the new mugs arrive. The stack gives you a studio. It does not give you a calendar.

Here the directory's trending strip is worth a look, because it is the closest thing on the page to "what are other people actually using". In the last seven days: Apache JMeter with 38 clicks, Code Composer Studio with 35, Apache AGE with 34, Stable Video Diffusion with 33, Amplitude with 33, TalkBerry with 32. Read that list as a shop owner. A load-testing tool, a code studio, a graph database extension, a video diffusion model, a product analytics platform, a voice AI. Not one of them is a social media tool. The trending strip measures clicks, not usefulness, and the clicks belong to developers — which makes sense, because the directory's own sponsorship page says its newsletter reaches 85,000-plus AI developers, a figure I am quoting as the demo's own claim. The directory knows its audience. Its audience is builders. I am a visitor in a builders' index, and the trending strip is the clearest proof of it. The week's posts got drafted in the end, and the directory's contribution was one image tool and a reminder that a media studio is not a marketing plan.

Job four: what the shop across the street charges

The shop across the street sells the same candles, and I needed to know whether my price was fair, high, or quietly losing me sales. This is a research job, so I went to the Research Lab Stack with genuine hope: Perplexity Pro does "live web and academic" search, DeepSeek V3 does "open MoE reasoning", CiteTrue verifies academic citations. A stack that can synthesise peer-reviewed papers should be able to tell me what a candle costs across the street.

It cannot, and the reason is the stack's own description: "synthesize peer-reviewed papers, extract structured insights, and verify factual assertions." Peer-reviewed papers. My competitor's price is not in a peer-reviewed paper. It is on a shelf across the street, or on their Instagram, or nowhere at all. The stack is built for a researcher whose question has citations. Mine has a price tag. Pointing a citation verifier at a candle price is using a microscope to read a shop window — the wrong instrument, however good it is.

So I did what shop owners have always done: I walked across the street, looked at the shelf, and came back. The price was fair. The directory contributed nothing to this job except the lesson, which is worth writing down because it is the morning's sharpest one: a directory's stacks are only as useful as the match between their persona and your job, and the Research Lab Stack's persona is a researcher, not a retailer. The directory never lies about this. Its stack descriptions are plain about who each stack is for. The failure is mine only in the sense that I kept hoping a stack would be for me. None of them is.

Job five: the newsletter nobody opens

Last job, and the one I most wanted to skip: the monthly newsletter. Two hundred subscribers, an open rate I am too embarrassed to print, and a blank page that has been blank since the first of the month. The job is to write something worth opening — three products, one story from the shop floor, one offer that is actually an offer.

The Solo Founder Stack is the directory's remaining answer, and it is the furthest from this job of all three: Cursor writes code, v0 generates interfaces, Claude 3.5 Sonnet does architecture logic, and the stack promises to take you "from natural language prototype to live production deployment with automated database and billing". A newsletter is not a production deployment. It is a letter. The stack is for someone building a product; I am someone who already has a shop and needs to tell two hundred people about it once a month. The catalogue, searched properly, has newsletter tools — of course it does, it has 33,379 tools — but by this point I had learned the directory's real shape: the front page shows you what is new, the stacks show you what three kinds of builder need, and everything else is behind a search box that only helps if you already know the name of what you want. I wrote the newsletter myself, in twenty minutes. The story was the Saturday chaos and the two-star review. The offer was ten percent off the new mugs.

The clock: what the morning actually cost

Here is the accounting, because a job test without a clock is an opinion, and I promised counting. The morning ran from nine to one — four hours, five jobs, one directory.

The review took about forty minutes, and six of those were writing the reply; the rest was reading the directory and learning that its front page does not do review replies. The ten descriptions took about an hour, forty minutes of writing and twenty of searching, and the directory's contribution was a writing tool found through search, not through any stack or front-page card. The week's social posts took about fifty minutes, and the directory contributed one image tool. The competitor's price took twenty minutes, and the directory contributed nothing — the walk across the street did it. The newsletter took twenty minutes, and the directory contributed nothing.

Add it up: roughly two and a half hours of real work, and roughly an hour and a half of reading a directory to find out what it could and could not do. Three of the five jobs got something from the directory, and two got nothing at all. And here is the honest limit I owe you, stated plainly: I did not sign up for, or run, any of the vendor tools behind the listings. This test is about what the directory itself shows — its stacks, its catalogue, its front page — and that is the whole point. A directory's job is to shorten the distance between a job and a tool. This one shortened it for two jobs, lengthened it for one, and stayed out of the way for two. It is the same "count first, then click" habit I used when I priced a real month of WhatsApp three ways: the sticker price is never the bill, and the front page is never the catalogue.

Would I use the directory again tomorrow? Yes — but differently. I would skip the front page entirely, skip the stacks entirely, and go straight to search with the name of the job in my hand. The directory is a good index wearing a bad shopfront: the stacks are for other people, the front page is four gift finders and ten copies of one sentence, and the search box is where the actual 33,379 tools live.

The verdict: who the stacks are actually for

So here is the whole morning in one paragraph, for the reader who skipped to the end. Jarvis's directory gives a small-business owner three curated stacks — one for a founder building software, one for a creator making media, one for a researcher reading papers — and none of them for a shop owner with five jobs before lunch. The front page offers twelve cards, four of them gift finders, ten of them sharing one identical description, and the trending strip measures developers' clicks, not a shop's needs. The directory still earns its place as a search box over 33,379 tools, and its comparison arena — Cursor at $20 a month for Pro, Copilot at $10, Claude Code by the token — remains the most useful table on the page, the same table the November 5 post in this series ran a full decision through. But as a guide for a small business, it is a map drawn for other travellers.

And the price of the script itself is honest, which matters if you are reading this as a builder rather than a buyer: the AI Tools Directory PHP SaaS Script with 24,671-plus pre-seeded tools is $37.00 for a regular license on codester.shop, instant ZIP. The gap between the shop's 24,671-plus seeded tools and the demo's 33,379 live tools is the same gap the November 5 and November 7 posts noted — the demo has grown past its seed count, and the two numbers measure different things. I would rather a seller understate the seed count than overstate it, and here the seller does.

If you run a small business and you open this directory tomorrow, do what I should have done at nine o'clock: ignore the stacks, ignore the front page, and search for the job, not the tool. Type "review reply", not "AI writer". Type "product description", not "content". And give yourself the same rule I gave myself for the descriptions: name one true thing before you allow one nice adjective. The directory will find you a tool. The judgement is still yours — and after this morning, I think that is exactly as it should be.

The habit behind this post is the same one behind the rest of this series. I vetted tools before signing up in the October 31 walkthrough, I built a week's tool stack from the directory's curated stacks as a founder and audited it for overlap and cost, I ran three coding assistants through the comparison arena on a twenty-dollar budget, and I counted what the Free label actually buys. The habit is the same every time: read the page as a buyer with a job, not a spectator with curiosity, and count what it actually tells you.

Frequently asked questions

What AI tools does a small business actually need?

Fewer than any directory will show you. This morning's five jobs — answering a review, writing product descriptions, drafting social posts, checking a competitor's price, sending a newsletter — needed one writing tool, one image tool, and a search engine. Everything else on the directory's front page, including its three curated stacks, was built for founders, creators and researchers, not for a shop owner.

Are Jarvis's curated stacks useful for a small business?

Mostly no, and the directory is honest about why: the Solo Founder Stack builds software products, the Content Production Stack makes broadcast-grade video and voice, and the Research Lab Stack synthesises peer-reviewed papers. None of those is a shop's job. The stacks are well-chosen for their own personas. A shop owner should skip them and go straight to the directory's search box.

How many AI tools are in the Jarvis directory?

33,379 verified AI tools across 27 categories, according to the demo's own figures at directory.tempomailusa.com, updated November 8, 2026. The shop listing for the script says 24,671-plus pre-seeded tools — the demo has grown past its seed count, and the two numbers measure different things (live index vs shipped seed data).

Why are four of the twelve front-page tools gift finders?

Because the front page shows the newest arrivals, not the most useful tools: all twelve cards were added a week ago, and four of them — What to Gift, Gifts AI, Present Finder and Giftly AI — are AI gift finders. A third of the freshest page is one job. If you need a gift finder you are spoiled for choice; if you need anything else, use search.

Why do ten tools share the same description?

The identical lines "AI-generated personalised plans" and "Recommendations from preferences" appear on ten of the twelve front-page cards. The directory ships with more than 24,671 pre-seeded tools, according to its shop listing, and seeded descriptions repeat wherever the seed put them. It is a data-quality seam, not a deception — but it means those ten descriptions tell you nothing about any of the ten tools.

How much does the Jarvis directory script cost?

$37.00 for a regular license on codester.shop, instant ZIP, verified live on 2026-11-09. That buys the script with its pre-seeded tools; it does not buy the tools listed inside it, and it does not change what any of those tools charge their own users.

What should a small business check before picking an AI tool from a directory?

Search for the job, not the tool — "review reply" rather than "AI writer". Read the stack descriptions as persona labels and skip any stack built for someone else. Check the comparison arena for a real price before you trust a catalogue label, the same habit the November 7 post used when it counted what the Free label actually buys. And the rule that saved this morning's descriptions: name one true thing about the product before you allow one nice adjective.