Guest Lecture · Business Computer Applications
AI in Business:
The landscape, tools, and techniques
DB
Darren Broemmer
Associate Professor · Computer Science
Overview
What We'll Cover Today
Five sections: from the current landscape to responsible use
01
The AI Landscape Where the industry stands: the gap between what's possible and what's in use
02
How AI Actually Works AI is based on patterns and math, not magic
03
Choosing Your AI Tools ChatGPT, Claude, Gemini, Copilot: what each does best and when to use it
04
Prompt Engineering How to ask AI for what you actually want, and build reusable templates
05
Critical Thinking with AI Automation bias, verification, and the two skills that matter most going forward
1
Section 01
The AI Landscape
Where the industry stands today
The AI Landscape
The Introduction of New Technologies
PERCEPTION
TIME
Peak Expectations
The Reality Gap
• Pilot failures
• Data quality issues
Pragmatic Adoption:
• Clear use-cases; targeted ROI
• Move from 'hype' to 'function'
Mature Technologies
• Integrated workflows
• AI as a utility
Concept inspired by Gartner’s Hype Cycle (Gartner, Inc., 1995–present).
The AI Landscape
Where Organizations Are Today
The gap between what workers use today and what's actually available
What Most Workers Use AI For Today
Ad-hoc chatbot queries — quick Q&A
Grammar and clarity edits on documents
Summarizing pasted content
Occasional copy-paste into reports
Basic research questions
Email drafting with light editing
What's Available Right Now
Complex reasoning across long documents
Multi-step workflow automation
Code generation and review
Research synthesis across sources
Document analysis and drafting pipelines
Agentic tasks: plans, executes, reports
Key insight: The professionals who understand how to close this gap will have a competitive advantage.
The AI Landscape
Where AI Actually Delivers Value
ISG State of Enterprise AI Adoption, 2025 — 400 senior AI decision-makers, $2.6B AI spend
AI OUTPERFORMS EXPECTATIONS
Regulatory Compliance +5.1% Product / Service Quality +4.1% Customer Satisfaction +2.6% Employee Satisfaction +2.0%
AI UNDERPERFORMS
Staff Reduction
−17.3%
vs. expectations
Key insight: AI excels at augmenting human work, not replacing it wholesale.
The AI Landscape
The Three Eras of Enterprise AI
How we got here and where we are headed
ERA 1
Up to 2022
Predictive AI
Pattern recognition, forecasting & anomaly detection from structured data. Built by data science teams.
Predictive maintenance · Risk scoring
ESTABLISHED
ERA 2
2022 – Present
Generative AI
Foundation models accessible to every employee. No data science expertise required, just prompting.
Copilots · Report drafting
WIDESPREAD USE
ERA 3
Emerging Now
Agentic AI
AI that plans, reasons across multiple steps, and orchestrates tools to complete complex workflows autonomously.
Multi-step project workflows
DESKTOP & SERVER
2
Section 02
How AI Actually Works
Patterns and math, not magic, and why confident is not the same as correct
How AI Works
What AI Actually Is
Not "Magic"
Sophisticated math applied to enormous amounts of text. Learned statistical patterns: not rules or logic.
Not Conscious
No understanding, intent, or awareness. It predicts the statistically likely next word.
Not Perfect
Fluent and confident, but structurally capable of generating false information.
At its core, every AI response follows this sequence:
Your Prompt
The input you provide
→
Tokenize
Break into numbered chunks
→
Find Patterns
Match training data
→
Predict
Generate likely output
→
Response
Streamed back to you
Key principles:
How AI Works
Two Things Every User Must Know
The Hallucination Problem
When AI generates plausible-sounding but false information
Why it happens: The training objective is to predict likely next tokens, not to retrieve verified facts.
Confident tone does NOT mean accurate output
Always review and verify AI-generated content
Treat every output as a first draft, not a final answer
Ask for citations, then verify those citations exist
The Jagged Frontier
AI capability is powerful yet unpredictable
AI may outperform an expert on a complex task, then fail at a simple adjacent one. The capability boundary is jagged, not smooth.
Task Complexity
Test AI on your specific tasks before relying on it
Plan for 6 months from now, not just today
AI capability has been doubling ~every 7 months
Source: Dell'Acqua, McFowland, Mollick et al. 'Navigating the Jagged Technological Frontier' — Harvard Business School WP 24-013 (2023)
3
Section 03
Choosing Your AI Tools
ChatGPT, Claude, Gemin, Copilot: what each does best and when to use it
Choosing Your Tools
The AI Tool Landscape
Microsoft Copilot
Microsoft 365 Integration
Context: 128K tokens
Lives inside Word, Excel, Outlook, Teams
Summarize meetings, draft emails
ChatGPT
Best General Purpose Tool
Context: 128K tokens
Excellent for brainstorming & Q&A
Most widely recognized and easy to start
Claude
RECOMMENDED
Best for Detailed Work
Context: 1M tokens
Highest writing quality and nuance
Analyze full documents & long context
Cowork: desktop agent automation
Google Gemini
Deep Research
Context: 1M tokens
Integrated with Google Workspace
Excellent multi-modal (text+image) and image generation
Choosing Your Tools
Choosing Your Approach: Task Duration × Integration
Standalone
Connected (Agentic)
Prompt & Copy
Any tool — standalone
Quick Q&A and explanations
Grammar and clarity edits
Brainstorming and idea generation
Summarizing pasted content
Connected Quick Tasks
All tools with integrations
Draft and send an email directly
Update a doc section or add a slide
Quick data lookups across files
Extended Dialogue
All tools
Multi-turn research sessions
Iterative report drafting
Back-and-forth problem solving
Long brainstorming threads
Agentic Workflows
Claude + Cowork
Agent plans and executes multi-step tasks
Reads specs, updates docs, sends summaries
HITL checkpoints at key decisions
Organize files into folders, manage email, presentation pipelines
4
Section 04
Prompt Engineering
How to ask AI for what you actually want and build reusable templates
Prompt Engineering
Be Specific: The Three Rules of Effective Prompting
The more context, the better the output
State requirements and constraints explicitly
Specify the process or steps to follow
Define output format (list, table, email, report)
Include length, tone, and audience
"Write a 3-paragraph executive summary of the attached report for a non-technical audience. Focus on budget implications and timeline."
Show, don't just tell
Provide 1-2 samples of what 'good' looks like
Even a rough example dramatically improves quality
Label clearly: "Here is a sample output:"
"Here is a sample action item: 'ACTION — Darren: Submit budget by Apr 30.' Format all items from this meeting notes doc this way."
Context shapes the AI's perspective
Tell AI who it is acting as
Domain expertise changes how AI approaches tasks
Roles leverage specialized language and frameworks
"You are a business analyst. Review this vendor proposal and identify the three biggest contract risks, citing specific clauses."
Prompt Engineering
From Prompts to Reusable Skills
A prompt library turns personal skill into institutional knowledge
1
Ad-Hoc Prompt
Typed fresh each time
Quality varies by person and day. No institutional knowledge. Fine for one-off tasks.
Example
"Draft a formal Request for Information. Be concise, cite the spec section, and state the needed clarification."
2
Saved Template
Tested & shared
A tested prompt saved with [PLACEHOLDERS] for project inputs. Consistent results anyone can run.
Example
"Review [SPEC SECTION] for compliance with [STANDARD]. Flag all missing items, conflicts, and ambiguities. Output as a numbered list."
3
Automated Skill
No coding required
A workflow that reads your files, runs the prompt, and writes outputs automatically. Connects to your documents and tools.
Example
"You are a project coordinator. Read the attached report. Extract all findings, flag deficiencies, and draft an email to the project manager."
5
Section 05
Critical Thinking with AI
Automation bias, verification, and the two skills that matter most
Critical Thinking with AI
Automation Bias: The Hidden Risk
Automation Bias: the tendency to over-rely on automated systems and accept their output without adequate scrutiny
Reliability Creates Its Own Risk
The more reliably an AI performs, the less critically humans scrutinize its output. Trust builds but oversight can erode.
Human Directs
→
AI Executes
→
Human Validates
Always keep a human in the loop for consequential decisions.
Novice Vulnerability Is Highest
Studies show novices exhibit significantly greater automation bias than experts. Experts calibrate trust over time through domain knowledge.
Sources: Georgetown CSET (2024); Spring / AI & Society (2025)
Critical Thinking with AI
The Two Skills That Matter Most
AI amplifies human skill — it doesn't replace the need for it
Skill 1: Domain Expertise
You have to know the subject to evaluate the output
1
AI as research partner: AI synthesizes fast, but you need domain knowledge to determine what's right, relevant, and missing.
2
Error detection: Hallucinations sound authoritative. Experts catch them. Novices trust them.
3
User responsibility: When you sign off on work, you own it: AI or not.
Skill 2: Critical Evaluation
Every output deserves a skeptical second look
1
Does it pass the smell test? Read AI output as a skeptic. If something seems off, dive into the details.
2
Verify key facts independently: Dates, names, statistics, citations: check them before using.
3
Calibrate trust by task: Highly reliable for grammar, summarization, etc. You need to determine how effective it is for your problem domain.
The takeaway: AI makes you faster, but your expertise and judgment are what make the output trustworthy.
Looking Ahead
Future Trends: What's Coming
These aren't distant possibilities — they're emerging right now
You Are Your Own "Mini-CEO"
Managing a fleet of AI agents to plan, research, draft, and execute.
Anyone can build software, automate workflows, or create customized tools without writing a line of code. The distinction between "user" and "developer" is dissolving fast.
Tokens Are the New Currency
Context windows are expanding rapidly, soon every AI will be able to hold entire knowledge bases in context. At least in the short term, tokens are the valuable resource.
Ethical Situations Are Everywhere
Transparency, accountability, integrity, and impact: these aren't abstract principles. Every time you use AI in your work, you're making an ethical decision about attribution, accuracy, and bias.
Key Takeaways
What to Remember
What to remember and put into practice starting today
1
The gap is your advantage Most workers use AI for basic tasks. Professionals who close the gap have a competitive edge.
2
AI is a pattern engine, not a fact machine Confident output is not the same as correct output. Always review and verify before using.
3
Pick the right tool for the task ChatGPT is your generalist. Claude is your specialist, use it for detailed reasoning and long documents.
4
Build your prompt library A tested, reusable prompt is a valuable asset. Start with 3–5 templates for your most common AI tasks.
5
Stay in the loop AI amplifies your human skills, it does not replace them. Domain knowledge + critical evaluation are the two skills that grow in value as AI advances.
Darren Broemmer · Associate Professor of Computer Science · Business Computer Applications