Module 1 · Lesson 1 of 21The Executive's AI BriefingFree preview

What AI Can Really Do Today

A clear-eyed tour of the four kinds of AI that matter to your business, what each can reliably do today, and how to tell real capability from hype in any meeting.

15 min read

Watch the overview first, then read the lesson.
On this page
  1. Learning objectives
  2. The four families of AI
  3. Capability versus hype
  4. Three questions that cut through hype
  5. Meet the companies in this course
  6. Try it
  7. Key takeaways

Every week brings a new AI announcement, and every vendor claims theirs will transform your business. Your board wants to know the plan. Your teams are already using AI tools, with or without permission. And you're expected to make investment decisions in a field where the ground shifts every few months.

You don't need to understand how the technology works under the hood to lead well. You do need three things: a way to sort what you're looking at, a realistic sense of what each kind of AI can reliably do today, and a few sharp questions that expose weak proposals fast. That's what this lesson gives you. It's the foundation for everything else in the course.

Learning objectives

  • Distinguish the four families of AI that matter to business: predictive machine learning, generative AI, automation, and AI agents
  • Describe what each family can reliably do today and where each is still immature
  • Use the capability reality check to classify any AI claim as ready now, ready with guardrails, emerging, or hype
  • Ask three questions that expose weak AI proposals in minutes

The four families of AI

"AI" is an umbrella term. In budget meetings, it gets used for everything from a spreadsheet macro to software that acts on its own. That's a problem, because these technologies have very different costs, risks, data needs, and maturity. Before you evaluate any AI proposal, sort it into one of four families.

Family What it does Typical business uses Maturity at the time of writing Main watch-out
Predictive machine learning Learns patterns from historical data to predict a number or a category Demand forecasting, churn risk, fraud detection, credit scoring, predictive maintenance, lead scoring Mature. Decades of production use Needs good historical data with known outcomes; accuracy drifts when the world changes
Generative AI Produces new content (text, images, code, audio) from plain-language instructions Drafting emails and proposals, summarizing documents, answering questions from your knowledge base, extracting data from messy documents, writing code Mature for drafting, summarizing, and extraction with human review; improving quickly Produces fluent, confident errors; quality depends heavily on the context you give it
Automation (rules-based) Follows fixed rules to move data and trigger steps, using workflow tools, integration platforms, or robotic process automation (RPA), which mimics clicks and keystrokes Routing approvals, syncing systems, generating routine reports, onboarding checklists Mature and predictable Brittle when inputs vary; can't exercise judgment
AI agents Combine a generative model with tools (search, business systems, email, code) to plan and carry out multistep tasks toward a goal Researching and drafting a prospect brief, resolving routine IT tickets, reconciling records across systems Emerging. Useful for narrow, well-defined tasks with checks; unreliable for open-ended work Errors compound across steps; needs tight permissions and monitoring

The most valuable solutions usually combine families. Picture an insurance broker that uses generative AI to pull data out of a PDF application, rules-based automation to push that data into its management system, and a predictive model to flag accounts likely to need a specialty insurer. The useful question is rarely "Which AI should we buy?" It's "Which combination fits this specific task?"

Tip: When someone says "we should use AI for this," ask which family they mean. The answer changes the cost, the timeline, the data requirement, and the risk profile.

Why generative AI put this on your agenda

Predictive models and automation have been in business use for years. What changed is generative AI, built on large language models (LLMs): systems trained on vast amounts of text that can read, write, summarize, and reason over language.

Language is the medium most knowledge work runs on. Emails, contracts, policies, call notes, proposals, and reports were hard for software to handle before. Now software can work with them, imperfectly but usefully. And anyone can use it by typing a request in plain language. That's why adoption has spread from the bottom up, often faster than IT departments could respond.

In practice, generative AI reaches your organization in two ways. The first is general-purpose assistants such as ChatGPT, Claude, Gemini, and Microsoft Copilot. The second is AI features built into software you already pay for: your CRM, help desk, office suite, and ERP. At the time of writing, most major business software vendors are adding generative AI features, often for an extra fee.

Capability versus hype

Every family has a frontier where marketing runs ahead of reality. Use the capability reality check to classify any claim you hear.

Category What it means Examples at the time of writing How to respond
Ready now Works reliably in production at many organizations with ordinary oversight Drafting and summarizing; meeting notes; extracting fields from standard documents; demand forecasting; first-line answers from a vetted knowledge base Move quickly. Focus on adoption and workflow fit, not on the technology
Ready with guardrails Works well, but errors are costly enough to require human review or tight constraints Customer-facing chat; contract clause review; claims triage; code generation Pilot with human review built in. Measure error rates before you expand
Emerging Impressive in demos and narrow deployments; inconsistent at scale Agents running multistep processes end to end; AI making final calls on complex judgment decisions Run contained experiments. Don't build a business case that depends on it yet
Hype Claims that outrun any evidence "Replace your whole finance team." "No data or process changes needed." "100% accurate." "Set it and forget it." Ask for evidence on your own data. Walk away from absolute claims

The boundaries move. Something "emerging" today may be "ready with guardrails" within a year, so revisit your classifications at least twice a year.

Watch out: Hype runs in both directions. Dismissing AI because a tool disappointed your team a year ago can be as costly as overbuying today. Capabilities that were unreliable not long ago are now routine. Re-test your assumptions on a regular cadence.

Three questions that cut through hype

You'll sit through many AI pitches, from vendors, consultants, and your own enthusiastic teams. These three questions separate serious proposals from wishful ones.

  1. What exactly is the task? "AI for customer service" is not a task. "Draft replies to billing questions using our policy library, for an agent to review and send" is. Vague proposals hide weak value.
  2. What happens when it's wrong? Every AI system makes mistakes. Who catches them, how, and what does an error cost? If the answer is "it won't be wrong," end the meeting politely.
  3. How will we know it's working? Ask for the baseline (how the task performs today) and the metric that will move. No baseline, no business case.

Example: A vendor pitches "AI-powered supply chain transformation" to Ironleaf Manufacturing, a fictional 400-person maker of industrial pumps. The CFO asks the three questions. The task: forecast demand for about 2,000 spare-part SKUs to cut both stockouts and excess inventory. When it's wrong: overstock ties up cash; a stockout delays a customer's repair and triggers expedite fees, so planners will review every forecast that falls outside normal ranges. How we'll know: compare forecast accuracy and inventory turns against the current spreadsheet method over two quarters. In ten minutes, a vague pitch became a specific, testable project. It also turned out to be predictive machine learning, not generative AI at all.

Meet the companies in this course

Throughout the course, you'll follow five composite organizations. They're fictional, built from patterns that recur across real businesses, and every number attached to them is illustrative.

Company Profile Where they are with AI
Calder & Voss Insurance Brokers Regional commercial insurance broker, about 180 people Account managers buried in submissions and re-keying data; competitors quoting faster
Ironleaf Manufacturing 400-person maker of industrial pumps and parts Strong ERP data, a skeptical workforce, and senior experts nearing retirement
Ashby Lane Advisors 120-person accounting and advisory firm Billable-hour model under pressure; partners split on AI
Westmere Health Network Regional healthcare network with clinics and two hospitals Heavy administrative burden and strict privacy obligations
Quillstack 300-person B2B software (SaaS) company Enthusiastic early adopters, scattered pilots, no clear ROI

You'll probably recognize your own organization in one or more of them. Notice how the same frameworks lead to different decisions in different contexts.

Try it

Spend about 10 minutes on this. It builds the habit you'll use for the rest of the course.

  1. List five AI claims you've heard in the last month: from vendors, the press, your board, or your own team.
  2. For each one, identify the family: predictive, generative, automation, agent, or a combination.
  3. Classify each with the capability reality check.
  4. Pick the claim most relevant to your organization and answer the three questions. Where you can't answer, note what you'd need to find out.
Claim:
Source:
AI family:
Reality check (ready now / ready with guardrails / emerging / hype):
1. The exact task:
2. What happens when it's wrong, and who catches it:
3. Baseline today, and the metric that would move:
What I'd need to find out:

If your organization has an approved AI assistant, paste in your list (without confidential details) and ask it to challenge your classifications. You'll see generative AI's strengths, such as fast, structured critique, alongside its limits: it doesn't know your context unless you tell it.

Key takeaways

  • "AI" covers four families (predictive machine learning, generative AI, automation, and agents) with very different maturity, costs, and risks. Sort before you evaluate.
  • Generative AI is the new ingredient because it works with language, the raw material of most knowledge work, and anyone can use it.
  • The best solutions often combine families. Ask which combination fits the task, not which AI to buy.
  • Use the capability reality check to separate what's ready now from what's emerging or hype, and revisit it regularly because the frontier moves.
  • Three questions expose weak proposals: What exactly is the task? What happens when it's wrong? How will we know it's working?

Enjoyed this preview?

Unlock all 21 lessons, 6 templates, and your certificate for $799.

Get full access