How AI is re-drawing the line between in-house, agency, and "done for you"
AI is changing who does the work in marketing. Here's how to decide what stays in-house, what goes to an agency, and when done-for-you makes sense.
TL; DR
- AI is collapsing the cost of execution across all three delivery models, shifting the real competitive advantage from who does the work to who designs the system that does it.
- In-house teams are evolving from "we do the work" to "we architect what gets done" — owning strategy, data, and AI infrastructure while delegating production to machines.
- Agencies that survive won't sell deliverables; they'll sell judgment, AI system configuration, and outcome-based retainers — the ones still billing by the hour for content production are in trouble.
- Done-for-you is moving from people executing on your behalf to platforms automating end-to-end — best for high-volume, repeatable tasks, but watch for genericness and vendor lock-in.
- The right question is no longer "can we afford to outsource this?" but "do we want to own this capability long-term?" — because what you outsource today may be a core competency you wish you'd built by 2027.
The "theory of the firm" asks a deceptively simple question: why do companies exist at all? Why not just contract everything out to the open market? The answer, first articulated by Ronald Coase in 1937, is that sometimes the cost of coordinating across markets is higher than the cost of just doing it yourself. Internal teams exist because they reduce what economists call "transaction costs."
For the past 30 years, that logic shaped how marketing got structured. Agencies existed because they had capabilities, headcount, and specialized tools that most brands couldn't afford to maintain internally. Done-for-you services existed because some tasks were repetitive enough to be packaged and sold wholesale. In-house teams existed to own the brand and coordinate the rest.
AI is rewriting Coase's math.
When a solo marketer can draft a campaign brief, generate 10 copy variations, optimize for SEO, and schedule distribution in an afternoon, the cost of doing it internally drops dramatically.
When an agency can deliver the same creative output with a fraction of the human hours, what exactly are you paying for? And when a platform can automate a lead nurturing sequence end-to-end, what does "done for you" even mean anymore?
The boundary between these three models isn't vanishing. It's moving. And for most teams, the old instincts about when to build, buy, or outsource no longer hold.
What these three models mean now
Before discussing where the line moves, it helps to be precise about what we're comparing. These three delivery models have always overlapped at the edges, and AI has made those edges fuzzier.
In-house teams
Traditionally, in-house meant internal employees doing the work: writing copy, running ads, managing social, building campaigns. The advantages were control, brand continuity, and institutional knowledge. The disadvantages were cost, specialization limits, and speed.
With AI, in-house looks different. According to Creative Review's survey of in-house agencies, 78% of internal creative teams are already using generative AI, primarily for concepting (70%), image creation (93%), and body copy (62%). These teams aren't getting bigger. They're getting faster.
The in-house model now means owning the brand, the data, and the AI infrastructure, while delegating more production work to AI tools that sit inside the firewall.
The shift is from "we do the work" to "we architect what gets done and by whom, including machines."
Agency-led work
Agencies built their value on three things: specialized talent, tools the client didn't have, and the capacity to scale up fast. AI threatens the first two directly. Specialized content production, ad optimization, SEO, and basic design are all areas where AI is compressing the skill premium.
According to Forrester report 2026: agencies are moving from building campaigns and flows to full-platform strategy, AI agent configuration, and outcome-based retainers. The ones that survive won't be selling "we write emails." They'll be selling "we build and operate the systems that drive your revenue lifecycle."
That distinction matters, and it requires a completely different kind of agency relationship.
Done-for-you services
Done-for-you has always been a spectrum. At one end, it means a consultant who handles everything with minimal client involvement. At the other, it means a managed service platform that delivers outputs automatically.
AI pushes DFY firmly toward the platform end. Services which handle conversational AI for property management and healthcare, don't just assist with tenant inquiries; they handle them end-to-end. That's not an agency executing manually behind the scenes. It's a productized system delivering outcomes. Human involvement is concentrated in setup, governance, and edge cases, not daily execution.
This is what DFY looks like now: not people doing the work for you, but systems doing the work for you, with people available when the system hits its limits.
Why AI is blurring the boundary
Three specific changes in AI capability explain why the old model boundaries are eroding.
The compression of production time
The most immediate effect of AI is speed. Tasks that took hours now take minutes. Tasks that required a specialist can be handled by a generalist with the right tools. PPC optimization on platforms like Google and Meta is already 90% automated by the platforms themselves. The question isn't whether AI can do this work; it's whether the people who used to do it can justify their fees.
For buyers, this means "we have capacity" is no longer a differentiator. Everyone has capacity now. The real question becomes: what do you bring beyond execution?
The shift from labor to orchestration
The real competitive edge in marketing is migrating from who does the work to who designs the system that does the work. Glen Slade, writing on LinkedIn about AI's effect on agency structures, makes the point that "AI agents do not simply pull work in-house; they reduce the friction of working with third parties." That's worth sitting with. AI doesn't kill outsourcing. It makes outsourcing easier, cheaper, and more modular.
What AI does kill is undifferentiated execution. If your value proposition is "we will do the thing," and AI can do the thing for a fraction of the cost, you have a pricing problem and a positioning problem simultaneously.
The accountability gap that's growing
As AI handles more of the work, questions of ownership and accountability get harder to answer. Who owns the output of an AI-generated campaign? What happens when an automated system makes a claim that turns out to be wrong?
Enterprise communications leaders at companies like Walmart and Dell are increasingly demanding transparency from agencies about which parts of deliverables are AI-generated, especially for investor-facing or high-stakes creative work.
This shapes how you structure the relationship, not just how you do the work.
The decision framework: what goes where
Most teams approach the in-house/agency/DFY question based on cost and capacity. That was the right lens before AI. Now, the more important questions are about control, compounding value, and risk.
The criticality test
If the work is directly tied to core brand positioning, proprietary customer data, or high-stakes decisions, investor communications, product strategy, pricing; it should stay in-house or at least be closely supervised by internal stakeholders. Not because AI can't help, but because the accountability needs to live somewhere internal.
A useful rule: if getting this wrong would embarrass your CEO or expose the company to legal risk, don't outsource the judgment. Outsource the drafting, maybe. Keep the judgment.
The repeatability test
Repetitive, standardized tasks are where DFY platforms win cleanly. Lead nurturing sequences, appointment scheduling, routine reporting, templated content, ad bid adjustments—these are predictable enough that automation adds almost pure value. The main risk is that "standardized" can slide into "generic" if there's no human check on brand voice and relevance.
The speed-to-market test
When you need something live quickly and don't have the internal bandwidth to do it well, agencies still have a strong case. Their value isn't just the talent; it's that they can start tomorrow, scale up fast, and absorb the project management overhead that internal teams routinely underestimate.
This is especially true for one-off campaigns, new channel experiments, or situations where you're entering unfamiliar territory and need someone who's done it before.
The compounding value test
If the capability you're building will matter more next year than it does today, and building it externally means you never own the institutional knowledge, that's an argument for keeping it in-house. Content strategies, brand voice, customer data systems, and AI infrastructure are all areas where internal ownership compounds over time in ways outsourcing never will.
This is where a lot of companies are currently getting it wrong. They're outsourcing things to agencies or DFY platforms that will become core competencies in two years, and paying to build expertise they'll never retain.
In-house vs. agency vs. done-for-you: the real trade-offs
The table above is a simplification, and real decisions rarely fit neatly into one column. Most mature marketing operations run some version of all three simultaneously: an internal team owns the strategy and brand, agencies handle specific channels or campaigns that require specialized depth, and DFY platforms manage repetitive execution. AI is changing the relative weight of each, not eliminating any of them.
How AI is changing what agencies sell
The most visible pressure point right now is on mid-market agencies: firms that built their business on content production, SEO, social media management, and PPC. These are exactly the services AI automates most directly.
The agencies navigating this well are doing a few specific things.
First, they're repositioning around judgment and implementation. As noted in the Senior Executive analysis of agency value, senior marketing leaders are shifting from buying outputs to buying "judgment, strategic intelligence, ethics, and implementation fluency." That's a different kind of service, priced differently, evaluated differently, and requiring a different kind of team.
Second, they're building AI infrastructure for clients, not just using AI internally.
The agencies that will thrive are the ones that can walk into a client's marketing operation, assess what's broken or missing at the systems level, and come back with a plan: which tools to use, how to configure them, how to train the AI on the client's brand voice, and how to measure whether it's working. That's consulting with a technical layer, and it commands a premium.
Third, they're moving toward outcome-based pricing. Billing for deliverables made sense when deliverables took hours of human labor. When a campaign brief takes 20 minutes with the right AI tools, billing by the hour or the piece becomes indefensible. Agencies that tie their fees to pipeline, revenue, or lifetime value metrics are protecting their margins and changing the nature of the client relationship.
Trust, disclosure, and the "AI voice" problem
There's a conversation happening in enterprise marketing that most smaller companies haven't caught up to yet: what are the disclosure obligations when you use AI in client deliverables?
Enterprise communications leaders at major companies are increasingly explicit about this. They want to know which elements of agency-delivered work are AI-generated, especially for high-stakes materials. Some are writing AI use requirements directly into SOWs. A few are using AI detection tools to spot-check claims that work is "human-written."
The more practical concern is quality. Overreliance on AI without strong human review produces what the industry has started calling "AI voice": content that's technically correct but tonally flat, predictable in structure, slightly generic in its specifics, and identifiable as AI-generated by anyone who reads a lot of it. When a client is paying for human creative expertise and receives AI voice, the trust damage is significant—and often hard to repair.
There are a few ways to address this in practice. The most obvious is human review at the end of every production workflow, not as a compliance step but as a genuine quality bar. The more sophisticated approach is what encoding brand voice, audience profiles, style guides, and product knowledge into the AI system itself, so the output sounds like the client before a human ever touches it. That kind of setup takes time and expertise to build, which is part of why it's a defensible service for agencies.
For buyers evaluating agencies and DFY providers, the right questions are: Where does human review happen in your process? What happens when the AI produces something off-brand? Who is accountable when something goes out with an error? Vague answers to any of these are a signal worth heeding.
Task-by-task: what AI handles, what humans own
Rather than speaking in generalities, it's worth being specific about which marketing tasks are shifting and in which direction.
Strategy and positioning: This stays human-led, with AI assisting research and synthesis. Brand strategy requires judgment about competitive context, customer psychology, and organizational trade-offs that AI handles poorly without extensive context. The mistake is using AI to write a positioning statement before you've done the thinking. Use it to pressure-test the thinking after.
Ideation and concepting: According to a survey, 73% of in-house creative teams use AI for generating options and visualizing different routes. This is a genuine win. AI is excellent at producing multiple variations quickly, which accelerates the creative selection process. The human task shifts from generating options to evaluating them.
Copywriting: Volume copy, product descriptions, email templates, social variations, SEO blog posts; is well-suited for AI with human review. High-stakes copy (homepage, sales narratives, brand manifestos) still requires human creative judgment and should only use AI as a drafting accelerant, not a final arbiter.
Design and visual assets: AI handles variation and iteration well. It's weak on original creative direction and often unreliable on brand consistency without extensive training on brand guidelines. Use it to multiply options; use humans to set direction and approve final work.
Analytics and reporting: AI is strong at generating reports, identifying anomalies, and summarizing data. The interpretation of what the data means and what to do about it is still human work—and it's where most marketing decisions get made.
Paid media optimization: Largely automated by the platforms themselves. The role of human experts, whether in-house or agency, is increasingly in strategy, audience segmentation, and creative testing frameworks, not daily optimization mechanics.
Customer-facing service and lead handling: Strong DFY AI territory for routine, high-volume interactions. Platforms like EliseAI demonstrate what's possible when AI handles end-to-end conversational workflows. The human layer is needed for complex cases, escalations, and relationship-critical moments.
How to evaluate an agency or DFY provider in the AI era
Buying agency or DFY services without asking about AI is like signing a construction contract without asking about materials. Here are the questions that matter:
- What AI tools do you use in your workflow, and at which stages? A vague answer here is a red flag. Good providers can walk you through the workflow step by step and explain exactly where AI assists and where humans take over.
- Who reviews AI-generated output before it goes to the client? If the answer is "we have a QA step," push further. What does that step involve? Who does it? What's the bar for passing?
- How do you handle brand voice and tone? Do they have a systematic way of training their AI tools on your brand guidelines, or are they relying on generic prompts and hoping for the best?
- What's your disclosure policy? If you ask them to mark which parts of a deliverable were AI-generated, will they do it? Discomfort or evasion here is information.
- Who owns the outputs and the model inputs? In a DFY arrangement where AI is trained on your data, make sure you understand who owns the configurations, the training data, and the outputs if you leave the platform.
- What happens when something goes wrong? How do they handle errors, off-brand content, or factual mistakes in AI-generated work? Is there a clear accountability path?
The team design question nobody's asking yet
Most of the conversation about AI and marketing delivery focuses on tools and cost savings. The more consequential question is about how your team is actually structured.
Companies that are figuring out AI-powered marketing tend to look different from traditional setups. They're leaner at the production layer — less time spent writing, formatting, and scheduling — and more focused at the strategic layer: deciding what to say, to whom, and why it matters. The job isn't disappearing. It's shifting toward people who can direct a system rather than operate one manually.
That shift is already happening for small teams and founders. Instead of hiring a content writer, a freelancer for SEO, an agency for campaigns, and a VA to keep it all moving, some are running all of it through a single system — one that handles the research, drafts, optimization, and distribution end to end.
Anjana Vasan
This is essentially what Tenet was built for. The platform does the heavy lifting across content, SEO, demand gen, product marketing, and social. And for teams that want someone accountable for actually shipping it week to week, Tenet Operator adds a dedicated person to the mix — one Operator who runs your marketing inside your account, reports back on what's working, and keeps the engine moving. No agency rotation. No freelancer juggling other clients. One person, one system, real accountability.
The point isn't that human judgment is going away. It's that the ratio is changing. A single Operator working with Tenet can handle what used to take a whole team — because the system handles the volume, and the person handles the direction.
For founders and small businesses, that's not a future state. It's available right now.
FAQ
What is the main difference between in-house, agency, and done-for-you in an AI context?
In-house means your team uses AI tools internally, owning the strategy, data, and brand governance. Agency means you hire an external team that uses both AI and human expertise to deliver work for you.
Done-for-you means a platform or managed service handles execution end-to-end, often with heavy automation and minimal involvement from your team. The practical difference is where accountability lives, how much customization you can get, and how quickly the capability compounds over time.
When does it make sense to keep AI work in-house?
Primarily when the work is tied to sensitive data, core brand positioning, or decisions with significant business consequences. Also when the capability you're building is strategic enough that you'll want to own the institutional knowledge long-term.
If a function will be central to how your business competes in two years, building it in-house, even if slower initially; usually creates more lasting value than outsourcing it.
When is an agency the better choice for AI-enabled marketing?
When you need fast deployment and don't have the internal expertise to do it well. When you need a type of specialization that's genuinely difficult to build internally—advanced data modeling, AI implementation, omnichannel CRM strategy.
When you're experimenting with a new channel or capability and want to test before committing to internal infrastructure. Agencies also make sense when you need a credible external perspective on strategy, not just execution support.
When does done-for-you work?
DFY works best for high-volume, repeatable workflows where the inputs and outputs are well-defined. Lead nurturing sequences, appointment scheduling, routine content production, ad optimization, and customer service at scale are all candidates.
The risks are genericness and vendor lock-in. If you go DFY, make sure you understand who owns the data and configurations, and that you have a plan for what happens if the vendor relationship ends.
How do I know if an agency is overusing AI or misrepresenting its use?
Ask directly, and be specific. Request a workflow map. Ask which tools are used at each stage. Ask to see examples of human review and QA processes. You can also use AI detection tools as a rough signal, though these aren't definitive.
The clearest signal is quality: generic structure, flat tone, oddly precise but contextually thin specifics, and predictable paragraph construction are all patterns associated with unreviewed AI output. If it reads like it could have been written for any client, it probably was.
How do we measure ROI across the three models?
The relevant dimensions are cost per outcome (not cost per deliverable), time to market, output quality (which requires defining what "good" means for your brand), and long-term use. In-house typically carries high upfront costs and lower marginal costs over time.
Agency costs are variable but carry relationship overhead. DFY has predictable subscription or managed service costs but may plateau on strategic complexity. Build a baseline before switching models, and measure against it consistently.
What tasks will always require human judgment, regardless of how good AI gets?
Anything where accountability matters and mistakes have real consequences. Brand positioning decisions, pricing strategy, stakeholder communication, complex customer relationships, ethical judgment calls, and crisis management all require human ownership.
The pattern is consistent: the more context-dependent, relational, or high-stakes the decision, the less AI can substitute for human judgment. AI can assist, draft, and accelerate all of these. It shouldn't be the final decision-maker.
