AI Agents are the new SaaS — Greg Isenberg

Source: YouTube | Channel: Greg Isenberg

AI Agents are the new SaaS. Software is moving from “help me do the work” to “do the work with me.”

Summary

Greg Isenberg presents a comprehensive playbook for building agent-first businesses, arguing that AI agents represent the next wave of value creation after SaaS. The key distinction: SaaS sells software tools, but Agent SaaS sells work—automating entire jobs rather than just providing productivity aids. The total addressable market is significantly larger because it addresses human capital, a multi-trillion dollar market.

Key Points

Core Mental Model

• SaaS sells software → Agent SaaS sells work
• Traditional SaaS: “Here is a tool your team can use”
• Agent SaaS: “Here is a job your team no longer has to do”
• Example: Restaurant phone agent that answers calls, handles reservations, routes VIPs, and integrates with OpenTable/Yelp

Finding the Right Opportunity

Start with workflows that have a paycheck attached. Good agent workflows have five traits:

1. High frequency — hourly is better than daily (every call, lead, ticket, quote)
2. Clear finish line — job booked, ticket categorized, refund approved
3. Touches existing software — Gmail, Slack, Shopify, HubSpot, ZenDesk, Stripe
4. Learnable edge cases — repetitive work with enough judgment that AI can help (not pure automation, not pure human judgment)
5. Buyer feels the loss — missed calls, slow replies, dropped leads, expensive humans doing low-value coordination

Before Building: Shadow the Human

• Watch someone do the job 10-20 times
• Ask for screen recordings with narration
• Understand: what makes a case easy vs. weird? What checks happen before decisions? Where do mistakes occur?
• The detail is the product — e.g., a restaurant host knows kitchen times, stroller-friendly tables, patio availability, VIP handling

Agent Specification (7 Key Parts)

1. What wakes the agent up?
2. What context does it need?
3. What tools can it use?
4. What is it allowed to do itself?
5. Where does it need approval?
6. When should it escalate to a human?
7. What does success look like?

Start Small: Minimal Useful Agent

Don’t aim for fully autonomous employee immediately. Four good first versions:

• Draft and approve — reads context, drafts reply/quote/summary, human approves
• Triage agent — classifies inbound work and routes to right place
• Coordinator agent — goes between systems and people, keeps work moving
• Bounded action agent — does specific thing under clear rules (e.g., process refund under $50)

“Many agent problems should start as workflows. Founders should earn autonomy by starting with a predictable path and adding judgment only when it creates value.” — Anthropic

The Product Wrapper Creates Trust

The agent does the work; the wrapper creates trust:

• Logs of what happened
• Approvals and controls
• Handoff rules
• Testing environment
• Explainability (why did the agent do this?)

Evals matter: Create a test set of 50 real examples (50 calls, 50 leads, 50 tickets). Run the agent against them to measure performance. This is also a powerful sales asset: “We tested this on your 50 old maintenance requests; got 42 right, flagged 6 for review, made 2 mistakes—and here’s how we fixed them.”

Go-to-Market Strategy

1. Sell the pilot like labor, then productize — start with 3 customers in one niche, same workflow, same pain
2. Pricing examples: 1,500 setup + 1,000/month per workflow; or 2,000 setup + 30 per qualified appointment
3. Distribution via workflow teardowns — show the painful old way (call comes in, nobody answers) vs. the agent way (call answered, questions asked, appointment booked, CRM updated, edge cases flagged)
4. Own one workflow — make the internet associate you with it through teardowns, benchmarks, memes

30-Day Launch Plan

• Day 1: Pick niche where missed work costs money (home services, property management, insurance)
• Day 2: Interview 10 operators, screen-share workflow research
• Day 3: Pick one workflow with frequency, pain, software access, clear success metric
• Day 4: Write agent spec (trigger, context, tools, rules, handoffs, eval)
• Day 5: Run manually with AI (Claude/ChatGPT) — copy-paste context, draft output, get human approval
• Day 6: Build smallest useful version (draft-and-approve or triage)
• Day 7: Create eval set from 50 real examples
• Week 2: Sell 2 pilots in same niche
• Week 3: Add product wrapper (logs, approvals, settings, analytics, handoffs)
• Week 4: Publish workflow teardowns, turn pilots into proof, double down on content strategy

Notes

• The opportunity is to find the smallest painful workflow that repeats all day in a niche you understand—and make it disappear
• Start with the job, shadow it, spec it, run manually, build smallest useful agent, sell pilot, then productize repeatable parts
• Build audience throughout; by end of month 4, have formats working and know where to spend paid ad dollars
• Focus on selling painkillers, not vitamins—managers/owners feel the pain of missed calls, slow replies, dropped leads

“Building agents is the new SaaS because software is moving from help me do the work to do the work with me.”