AI agent business models: how to build a sustainable agency

McKinsey’s State of AI report shows that AI-native services are growing at 3x the rate of traditional consulting, validating the shift toward agent-building as a sustainable business model.

TL;DR: After trying 5 business models over 9 months, the winning approach is a hybrid: fixed-price agent builds ($1,400/14 days) for portfolio and cash flow, plus maintenance retainers ($500-2,000/month) for recurring revenue. Skip SaaS and revenue share as a solo developer — the $25/hour build work funds the $212/hour retainer work.

When I started Agentic Up, I had one question: how do you charge for something this new?

AI agents aren’t products (fixed scope) and aren’t consulting (variable scope). They’re somewhere in between — you build something specific, but the boundaries shift as LLM capabilities evolve weekly.

I tried five business models over nine months. Some worked. Some nearly bankrupted me.

Key takeaways:

• Fixed-price per agent ($1,400/14 days) works well for vertical, well-scoped agents
• Hourly consulting ($150-200/hr) pays the bills but doesn’t scale
• Retainer/maintenance ($500-2,000/mo) is where the real recurring revenue lives
• SaaS subscription and revenue share are tempting but impractical for a solo builder
• The winning model is a hybrid: fixed-price build + retainer for ongoing maintenance

Model 1: Fixed-price per agent ($1,400/14 days)

This became my primary offering. Build a vertical agent in 14 days for a fixed price.

The pitch: “You describe the workflow. I build the agent. 14 days. Fixed price. You own the code.”

Revenue per project: $1,400 × ~2 projects/month = $2,800/month at peak

How it works:

• Discovery call (1 hour) — understand the workflow
• Scope document — exactly what the agent does, how it’s deployed, what success looks like
• Build week 1 — agent loop, tools, core functionality
• Build week 2 — testing, deployment, documentation
• Handoff — code + deployment + 30-minute walkthrough

What went right:

• Clear expectations. Both sides know what $1,400 buys
• Forces scope discipline. I can’t build general-purpose agents in 14 days, so I don’t try
• Builds portfolio quickly. I have 12 case studies now

What went wrong:

• Inbound volume is inconsistent — 0-4 leads per month
• Some agents genuinely need more than 14 days. I lose money on those
• Post-delivery support bleeds into unpaid time

The math:

Item	Cost
LLM API costs during development	$50-100
Deployment hosting (Railway, first month)	$5-20
My time (14 days × ~4 hours/day)	56 hours
Effective hourly rate	$1,400 / 56 = $25/hour

$25/hour is decent for India but not great. The leverage comes from reuse — the 10th agent takes half the time of the first because I reuse the scaffolding.

Model 2: Hourly consulting ($150-200/hr)

The pitch: “I’ll build whatever you need, billed weekly.”

Revenue: $150-200/hr × ~20 billable hours/week = $3,000-4,000/month

What went right:

• Simple. I track time, send invoice, get paid
• Works for clients who don’t know what they want and need exploration
• Highest hourly rate of any model

What went wrong:

• Not scalable. 20 billable hours is my max while maintaining quality
• Clients question time. “Why did it take 3 hours to set up the API?”
• No leverage — every hour is traded for dollars
• Feast-or-famine. Some weeks are 30 hours, some are 5

Verdict: Good for cash flow, terrible for building a business. I use this sparingly for existing clients who need urgent work.

Model 3: Retainer/maintenance ($500-2,000/month)

The pitch: “I built your agent. It runs. But APIs change, LLMs improve, and things break. Let me keep it running.”

Revenue: $500-2,000/month × 5-10 clients = $2,500-20,000/month

What went right:

• Recurring. I know how much I’ll make next month
• Low effort. Most agents need 1-4 hours/month of maintenance
• Deep relationships. Retainer clients become long-term partners
• The 60% conversion rate from build → retainer is real

What went wrong:

• Hard to get clients to commit upfront. They want to “see how it goes”
• Scope creep — “while you’re maintaining the agent, could you also…”
• Some agents truly don’t need maintenance after initial setup

My retainer structure:

Tier	Price	What they get
Basic	$500/mo	Bug fixes, API updates, monitoring uptime
Standard	$1,000/mo	Above + monthly improvements, performance tuning
Premium	$2,000/mo	Above + priority support (4hr response), new features

Actual conversion data: Of 12 agents built, 7 are on retainer (58%). Average retainer: $850/month. Monthly recurring revenue: $5,950.

Model 4: SaaS subscription ($49-199/seat/month)

The pitch: “Use my agent platform. Pay per seat.”

Revenue potential: High, if you can build a product, not a service.

What went right:

• Nothing, because I didn’t execute this well

What went wrong:

• Building a SaaS product as a solo developer is a multi-month commitment with no revenue upfront
• AI agent SaaS is already crowded. Standing out requires either deep vertical expertise or significant distribution
• Support burden for SaaS is higher than agency work — you need documentation, onboarding, support channels
• The economics don’t work until you have 50+ paying users

The math:

SaaS Tiers	Monthly Price	Users to Match Agency Revenue	Required MRR
Basic	$49	82	$4,018
Pro	$99	40	$3,960
Enterprise	$199	20	$3,980

To match my agency revenue (~$8,000/month), I’d need 80+ paying users at $99/seat. That’s a lot of marketing and sales for a solo developer who’s also building the product.

Verdict: I haven’t made this work yet. But I see the potential. If I were starting over with more runway, I’d explore a vertical SaaS — one agent that solves one problem for one industry — rather than a general-purpose agent platform.

Model 5: Revenue share / equity

The pitch: “I’ll build your agent for free / reduced rate in exchange for 5-10% of revenue or equity.”

Revenue potential: Unlimited (if the client succeeds) or zero (if they don’t).

What went right:

• One deal worked: a real estate agent that generates listing descriptions. They pay me 5% of the revenue the agent generates. Current monthly: about ₹41,000 ($500).
• Aligned incentives — we both want the agent to succeed

What went wrong:

• Most startups fail. I’ve done 3 revenue-share deals. One pays. Two paid nothing.
• Difficult to verify revenue. I trust my clients, but it’s not auditable
• Delayed payment — revenue share pays monthly, often starting small
• Legal overhead — contracts are more complex than fixed-price

Verdict: Only do this if: (a) you believe in the client’s business, (b) you can afford to work for deferred compensation, (c) you have a clear measurement and payment mechanism.

What I recommend

After nine months, here’s the model I’d recommend for a solo AI agent developer:

Phase 1: Fixed-price builds (months 1-6)

• Charge $1,400-2,000 per agent
• Focus on 14-day delivery with tight scope
• Build case studies and a repeatable process
• Target: 2 builds/month = $2,800-4,000/month

Phase 2: Add retainers (months 3-12)

• Convert 60% of build clients to maintenance retainers
• Charge $500-2,000/month per client
• Target: 8 retainers at avg $850 = $6,800/month MRR
• Combined with 1-2 builds/month: $8,000-10,000/month

Phase 3 (optional): Vertical SaaS

• Only after Phase 2 provides stable income
• Pick the most repeatable agent from your builds
• Turn it into a self-serve product
• Target: 50 paying users at $99/seat = $4,950/month

When to say no

These client red flags cost me time and money:

1. “Build me an AI thing” — no scope, no clear problem. Walk away.
2. “We need a general-purpose AI assistant” — those never work. Agents need focused, narrow use cases.
3. “We’ll pay you after it generates revenue” — risk-shifting disguised as partnership. Hard pass.
4. “We need you to sign an IP assignment for anything you build” — they want your methodology, not your agent. No.
5. “This is simple, should take a weekend” — it’s never simple. This client undervalues the work.

The bottom line

My current monthly revenue breakdown as of June 2026:

Stream	Monthly Revenue	Monthly Hours	Effective Hourly
Fixed-price builds	$3,500 (2.5 avg)	140	$25
Retainers	$5,950 (7 clients)	28	$212
Hourly consulting	$1,500 (sporadic)	10	$150
Revenue share	$500 (1 deal)	2	$250
Total	$11,450	180	$63 avg

The $25/hour on builds funds my portfolio and feeds the retainer pipeline. The $212/hour on retainers pays my rent. The revenue share is a lottery ticket.

If you’re starting today, skip SaaS. Skip revenue share. Start with fixed-price, build case studies, and convert to retainers. That’s the path that actually works.

---

Related: AI agent pricing: how much to charge for custom agents — a transparent breakdown of pricing custom AI agent projects. Also see AI agent pricing strategies 2026 for a comprehensive pricing guide.