AI Agency vs. Hiring In-House: Which One Actually Makes Sense?

Key Points

• In-house AI engineering costs $280K-320K in year one when accounting for salary, benefits, equipment, onboarding delays, and reduced early productivity—then $180K-200K in productive value in year two as the engineer ramps up.
• Agencies have lower upfront costs and no ramp-up period, but ongoing projects cost more per feature than in-house development after three years, making each approach optimal at different company growth stages.
• Choose in-house if you need continuous AI work for multiple years with growing complexity and institutional knowledge; choose an agency if you're starting out, have uncertain scope, or need specialized expertise without permanent headcount.

This is the question I get asked at least once a week: "Should we just hire someone in-house instead of using an agency?"

The answer is always: "Maybe. Let me explain why."

I'm going to be unusually honest here, because I run an agency but I genuinely believe in-house is the right choice for some companies. And I'd rather you make a good decision for your business than make a bad decision that lines my pockets. Bad decisions make for bad clients anyway.

Let's break down both paths with actual numbers and real trade-offs.

The In-House Math (It's More Than Just Salary)

When people think about hiring an in-house AI engineer, they do the math wrong.

What they think it costs:

• Salary: $150K/year for a solid engineer
• Total: $150K

What it actually costs:

• Salary: $150K
• Taxes and benefits (30%): $45K
• Equipment, software licenses, workspace: $5K
• AWS/cloud infrastructure: $10K-30K/year
• Training and keeping them current: $3K-5K/year
• Year 1 total: $213K-233K

But that's not even the real cost.

The hidden costs:

• Onboarding time: A good engineer isn't productive for 3-6 months. They're learning your codebase, your infrastructure, your business logic, your data architecture. During this time, you're paying full salary for maybe 40-50% productivity.
• Ramp-up delay: Want them doing real work? Add another 2-3 months.
• Context switching: They're not just doing AI work. They're answering questions, sitting in meetings, context-switching between projects. Real productive time is maybe 60-70% of their calendar.

Real year-one cost when you factor in these things: $280K-320K for actual deliverables.

Compare that to year two: they're fully productive, they own systems, they iterate on previous work. Year two actual cost: maybe $180K-200K in real value delivered.

Three-year cost: $580K-720K total investment before they're really humming.

The Agency Math (It's Simpler But Different)

When you work with an agency like Rotate, you typically:

• Pay per project: $5K-25K depending on scope
• Average timeline: 4-8 weeks per project
• Get delivery: working code, documentation, ownership

Year one cost: Let's say you do 2-3 projects. That's $30K-75K out of pocket.

What you get: Working solutions, code you own, zero maintenance burden (unless you want it).

Year two: If you run 3-4 more projects, that's another $45K-100K.

Three-year cost: $120K-250K depending on project volume.

On the surface, that looks cheaper. And it is. But here's the catch: you don't own the talent. The agency can't be on retainer for internal problems. You get point solutions, not a continuous system building capacity.

The Real Decision Framework

Here's what actually matters:

Choose In-House If:

1. You have enough AI work to keep someone busy full-time

This is the big one. A full-time engineer needs about 4-6 substantial projects per year to stay engaged and productive. If you only have 1-2 projects planned, you're paying a lot for idle time.

Most mid-market companies? They have 2-3 AI projects per year. That's not enough to justify a full-time hire.

2. You have AI as a core part of your business

If AI is fundamental to how you operate — if your product is AI, if your revenue depends on AI — then you need in-house ownership. You need someone who lives and breathes your systems. You need continuous iteration, maintenance, and improvement. That's not a project; that's a business function.

3. You have the infrastructure to support them

Do you have cloud infrastructure set up? Do you have data pipelines? Do you have a development process? Do you have a senior engineer who can mentor them? If you're starting from nothing, add 3-6 months to ramp time (and another $50K in infrastructure work).

4. You can afford the ramp-up cost

You need to be able to absorb 6-9 months of low productivity while they learn your business. Most small companies can't. If you've got 3 critical AI projects waiting, you can't wait 9 months for someone to be useful.

5. You have the risk tolerance

What happens if they leave? What happens if they're not as good as you thought? What happens if the technology they built on becomes obsolete? In-house means you carry that risk.

Choose an Agency If:

1. You have fewer than 4 projects per year

You don't have enough work to justify a full-time headcount. An agency lets you pay for what you use.

2. You want delivery speed

We've built the same integrations 50 times. We know the pitfalls, the shortcuts, the gotchas. We build faster. Your in-house engineer is learning your specific systems from scratch.

On our first project with a new client, we're maybe 10% slower than we'd be with our own systems. On your in-house engineer's first project at a new company? They're 50% slower because they're learning everything.

3. You want to avoid hiring risk

Hiring an AI engineer is genuinely hard. You need someone who's technical, understands ML concepts, can work with real data, and can communicate with your business stakeholders. That person is scarce and expensive.

When you hire the wrong person, it's a 6+ month problem that costs you six months of salary plus the cost of hiring again.

4. You don't have infrastructure yet

If you don't have a mature engineering organization, an in-house engineer will spend the first 6 months setting up infrastructure instead of building AI solutions. An agency comes with that knowledge and can move faster.

5. You want to keep things simple

An agency relationship is transactional. You scope work, we deliver, you pay, done. Your in-house hire is a long-term relationship with all the complexity that entails. Taxes, benefits, management, growth expectations, career development. Most business owners don't think about this, but it's real.

The Hybrid Approach (And Why It's Harder Than It Sounds)

Some companies try: hire an in-house engineer and also use an agency. Theory: you get continuous capacity plus burst capacity for big projects.

Reality: This is hard to manage.

Your in-house person gets frustrated because the agency is making more per hour than they are. Your agency gets frustrated because your in-house person is fighting change and blocking decisions. They have different incentives, different communication styles, different technical cultures.

It can work, but you need a senior person on staff who can manage that relationship and translate between cultures. Most companies don't have that, so the hybrid becomes a chaos situation.

The Numbers in a Real Scenario

Let's say you're a $5M revenue company. You have three AI projects on your roadmap:

Project A: Lead scoring automation (estimate: $12K, 6 weeks) Project B: Customer segmentation (estimate: $15K, 8 weeks) Project C: Internal process automation (estimate: $8K, 4 weeks)

Total work: ~18 weeks, $35K

Path 1: Hire an engineer

• Year 1 cost: $280K (salary + hidden costs + no productivity ramp)
• You get: Maybe Project A done, maybe a start on Project B
• You also get: Someone who can do maintenance, iterate, and handle new problems as they arise

Path 2: Use an agency

• Year 1 cost: $35K
• You get: All three projects done in 18 weeks
• You also get: Code you own, documentation, zero hiring risk
• You don't get: Continuous capacity for unexpected requests

Path 3: Hybrid (one engineer + agency for big projects)

• Year 1 cost: $280K (engineer) + maybe $15K (agency for Project C)
• You get: Engineer owns Projects A and B, agency does C
• You have to manage: integration between two different approaches

In this scenario, the agency wins on pure economics. But if you had 6-8 projects per year instead of 3, the engineer wins.

What Actually Tips the Scales

Here's the real decision maker that nobody talks about:

Do you want a business function or a talent asset?

If you want to build AI capabilities into your organization, and you want people who know how to do this, you need in-house. But that's a 2-3 year investment, and you need enough work to justify it.

If you want to solve specific problems fast and own the solutions, you want an agency.

Most companies at the $2M-10M revenue level land in the middle: you have enough AI opportunities to make it real, but not enough to justify a full-time dedicated engineer. For you, an agency + maybe one technical person on staff (to manage vendors, understand what you're buying, maintain what gets built) is the right answer.

By the time you're at $50M+ revenue and AI is core to what you do, you hire in-house and scale from there.

One More Thing: The Quality Question

I'd be remiss if I didn't say this: not all agencies are the same, and not all in-house engineers are the same.

A bad in-house engineer costs you way more than a bad agency because you're stuck with them for months. A bad agency costs you less directly, but you get bad code you have to maintain forever.

A good in-house engineer owns the problem end-to-end and iterates continuously. A good agency owns the project, delivers clean code, then moves on.

The question isn't just "agency vs. in-house." It's "good agency vs. bad agency vs. good engineer vs. bad engineer."

If you're going to hire in-house, make sure you can actually assess technical quality and you have someone senior who can mentor and manage them. If you're going to use an agency, make sure they give you ownership of the code and actually document what they build.

The Honest Take

If you have 3-4 AI projects planned over the next year, use an agency. You'll save money, move faster, and reduce hiring risk.

If you have 6+ projects per year and AI is becoming core to your business, start planning for an in-house hire. But hire slowly, invest in onboarding, and maybe use an agency for your biggest projects while they ramp.

If you're somewhere in between, you're probably overthinking it. Start with a couple of agency projects, prove out the solutions, and see if you actually have enough work. Then hire.

Don't hire prematurely. I see companies bring on a full-time engineer for "AI stuff" and then give them 3 projects' worth of work per year. They get bored, they leave, and you're stuck hiring again. It's not worth it.

Want to figure out which path makes sense for your business? Let's talk. I can help you think through the decision, even if it means recommending you hire in-house instead of working with us.