Siah Labs×AACCNJ
Presenter

Workshop · AACCNJ

AI for Main Street

A workshop to understand AI tools and agents, spot the traps, and leave with moves you can use this week.

Hosted for African American Chamber of Commerce of New Jersey. Leave with clarity, not buzzwords.

Resources

Open the tools

Free accounts work fine for tonight. Open one tab per tool, then come back here for prompts and demos.

Quick wins

What should I do?

Every actionable tip from the deck and this hub, in one place. Skip the stories if you want - read this and you are good to go.

01

Before tools

Start with the task

  1. 1

    Name one painful weekly job before you open a product list.

    “Which AI should I buy?” is the wrong first question.

  2. 2

    Write the steps of that job on paper - trigger, steps, done.

    That packet is day-one training for any smart intern.

  3. 3

    Ask: What’s the destination? What’s the vehicle? What’s AI’s role?

    A real outcome, the right system, and clear access, knowledge, and “done.”

02

Choose the vehicle

Three questions before you pick a tool

  1. 1

    One-time task → use AI chat.

    Think, learn, write, brainstorm. Fast and flexible.

  2. 2

    Repetitive work → use a workflow.

    Connect apps, kill copy/paste, run the same path every time.

  3. 3

    Critical to the business → use software with controls.

    Permissions, reporting, customers, and money need a real system.

    See the spectrum →
03

Stay in charge

Rules that keep you out of trouble

  1. 1

    Treat every AI output as a draft until you approve it.

    Fluent answers feel finished even when they are wrong.

  2. 2

    Put a child lock on irreversible actions: send, pay, delete, post.

    Typing “DO NOT” louder is not a safety plan. Hard stops live outside the model.

    Child lock explainer →
  3. 3

    Verify important facts against your records; check math in a spreadsheet.

    Ask for sources. Show the steps. Never trust a confident total alone.

  4. 4

    Don’t paste SSNs, bank details, secret contracts, or anything you wouldn’t email a stranger.

    When in doubt, redact. Prefer tools connected to your data with clear rules.

  5. 5

    Paste current prices, hours, and examples - keep a human on tone and relationships.

    Missing context and stale facts are how good drafts go wrong in public.

04

First project

A good starting point has six signals

  1. 1

    Frequent - happens weekly or daily.

    Rare edge cases rarely pay off first.

  2. 2

    Meaningful - consumes real time or attention.

    Skip trivial annoyances.

  3. 3

    Repeatable - the usual steps can be written down.

    If the process changes every time, wait.

  4. 4

    Data ready - needed information is accessible and reasonably clean.

    Facts trapped in people’s heads stall agents.

  5. 5

    Reversible - mistakes can be caught, corrected, or held for approval.

    One wrong move with lasting harm needs a lock first.

  6. 6

    Measurable - you can see time or outcome improve.

    Start small, reversible, and measurable.

    Open opportunity canvas →
05

This week

Four moves sized for busy owners

  1. 1

    Tonight: Pick one repetitive task you do weekly and write the steps on paper.

    Clarity beats tools. AI only helps once the work is visible.

  2. 2

    Tomorrow: Open the prompt library and run 3 prompts on a real email, meeting note, or spreadsheet.

    Muscle memory forms faster on your own work than on demos.

    Open prompt library →
  3. 3

    This week: Create a “verify before send” rule for anything customer- or money-facing.

    Confidence without verification is how AI quietly creates risk.

  4. 4

    Next Monday: Use the opportunity canvas scorecard on one candidate task - frequency, data, risk, measurability.

    A small, reversible, measurable win is the best place to begin.

    Open opportunity canvas →

Your 2 hours

The path through the room

Jump to any block. Demos and hands-on sessions are where the lightbulbs usually go off.

  1. 0:00
    How we got hereTeach

    From human computers to pocket supercomputers, and what that means for your work.

  2. 0:15
    The smart intern modelTeach

    What AI is great at, where it fails, and how to stay in charge.

  3. 0:35
    Live demos for NJ businessesDemo

    Real prompts plus an end-to-end agent using tools, approval, and an activity log.

  4. 1:00
    Chat vs workflow vs softwareTeach

    Pick the right level of control for the job, not the hype.

  5. 1:20
    Hands-on: map one opportunityHands-on

    Define the job, tools, decisions, child locks, and success measure for one real workflow.

  6. 1:45
    Decision framework + takeawaysDecide

    Three questions before you pick a tool, plus this-week moves you can take home.

Understand the limits

AI is incredible. It is not magic.

Tap each failure mode. This is the section people quote when they say they finally get what these tools can and cannot do.

Where AI fails

Hallucinations

It can invent names, dates, policies, or sources with total confidence.

In the room

·

CPA / Bookkeeper

“What NJ tax credit can my salon client claim for hiring two apprentices?”

AI said

Names the “Garden State Apprentice Hiring Credit,” cites Form GST-441, and gives an Oct 15 filing deadline.

Reality

That credit does not exist. The form number is made up. It sounded official enough to forward to a client.

What to do instead

Ask for sources. Verify anything important against your records.

The child lock

The fix isn’t typing louder

“Do NOT do XYZ” is asking for good behavior. A child lock removes the option - the unsafe action is blocked outside the AI’s control.

Behavior

  • Louder instructions in the prompt
  • Usually works, not always
  • Still a judgment call every time

Child lock

  • Human approves before send or save
  • Irreversible actions blocked by design
  • Hard stop the model cannot override

What happens if this goes wrong at 2am on a Saturday, when nobody is watching?

Plain English

Words you will hear tonight

Tap a term. No jargon required - just enough vocabulary to stay in charge of the tools.

Plain English

Model

The AI “brain” trained on lots of text. ChatGPT, Claude, and Gemini are products built on models.

Main Street: Like hiring a very fast intern who has read a lot - but has never worked at your company.

Demonstrations

Main Street problems, live

Dummy walkthroughs for now. In the room, we run these on real documents from the audience.

Prompt

Catch jobs that are slipping

Here is my job list with status and last update. Which jobs are falling behind, and what should I ask each crew lead tomorrow morning?

What you get

Flags 3 late jobs, drafts a short check-in for each crew, and suggests which customer to call first.

Live during the workshop. Try the same pattern on your own data in the hands-on block.

From prompt to agent

Watch the whole job move

A prompt produces an answer. An agent pursues an outcome across multiple steps, uses tools, stops at the child lock, and logs what happened.

Incoming inquiry

Website redesign inquiry

Dana · Garden State Dental

We need a new website before our second location opens. We are in Montclair and hope to launch in about eight weeks. Can your team help, and what would the next step be?

Agent run

Lead follow-up

Start the run, then advance one stage at a time. Watch where the agent uses tools, makes a decision, and has to stop for a human.

Prompt library · 40 prompts

Run them live. Right here.

Filter by job, paste your real material, and run any prompt against a live model without leaving this page - or copy it into ChatGPT or Claude to keep using it later.

  • EmailAny

    Rewrite a stiff email

    Rewrite this email so it sounds like a capable small-business owner: clear, warm, and short. Keep the facts. Remove corporate fluff.
    
    [PASTE EMAIL]

    Tone is the first thing customers notice - and the easiest win.

  • EmailServices

    Apology that rebuilds trust

    Draft a short apology email for a late delivery. Own the mistake, say what we are doing next, and offer one concrete make-good. No over-apologizing.
    
    Context: [WHAT WENT WRONG]

    Structure beats emotion when you are under pressure.

  • EmailSales

    Follow up without being pushy

    Write a 4-sentence follow-up to someone who asked for a quote two weeks ago and went quiet. Helpful, not needy. Include one soft next step.
    
    Original ask: [CONTEXT]

    Most lost deals die from silence, not from a bad offer.

  • EmailAny

    Meeting notes to action list

    Turn these messy notes into: (1) a 5-bullet summary, (2) decisions made, (3) open questions, (4) action items with owners.
    
    [PASTE NOTES]

    Capture the meeting once; reuse the structure forever.

  • EmailAny

    Decline politely

    Write a polite no to this request. Leave the door open for future work. Two short paragraphs max.
    
    Request: [PASTE]

    A clean no protects your calendar and your reputation.

  • EmailOps

    Vendor status chase

    Draft a firm but professional email asking for a status update on a late order. Ask for a date, not a vague assurance.
    
    Order details: [PASTE]

    Specific asks get specific answers.

Visualization

There is no best AI

Imagine hiring a brilliant assistant - then forgetting to give him a car, a desk, or access to your systems. The assistant is only as effective as the system around it. Slide across the spectrum and choose based on the job.

FlexibleControlled

ChatGPT

Fast thinking partner. Great for drafts, research, and learning.

Best for

  • Brainstorming
  • First drafts
  • Explaining concepts

Watch outs

  • Manual copy/paste
  • Inconsistent outputs
  • Easy to overtrust

What custom software looks like

Built tools you can touch

Proof at the controlled end of the spectrum. Some use AI in the loop. One is pure software - built with AI, delivered without the chat.

Snap a card at a mixer. Review. Into the CRM.

Photo in, structured contact out. The human review step is the child lock - nothing hits the CRM without approval.

AI in the loop
  1. 01Photo
  2. 02Extract
  3. 03Review
  4. 04Saved

A card from tonight’s mixer — tap Extract to run the vision pass.

Input

Keisha Morgan

Ops Manager

Harbor Street Catering

(973) 555-0142

[email protected]

Newark, NJ

Output

Waiting for extraction…

Hands-on · 25 minutes

Map one opportunity worth building

Work through the job, the agent, and the child locks. Then score the idea, get a live AI review of your plan, and leave with one specific next action.

Private to this browser. Nothing is submitted.

0-4 min

Name one job worth improving

Pick something real that happens repeatedly. One clear opportunity beats three vague ideas.

Take home

Take the whole workshop home

Jump back into any block, or open the printable recap to save as PDF from your browser.

Included for attendees
Stay connected

Get the recap + prompts emailed to you. No spam - workshop follow-up only.