For years, “productizing” a service was a pricing decision for freelancers and small agencies. In 2026, Y Combinator and Sequoia elevated it to an investment thesis. These leading funds argue that services firms that use artificial intelligence (AI) to sell finished work at a fixed price could have margins as attractive as a SaaS company’s. Is that true?
“Services are the new software”
In February 2026, Y Combinator (YC), Silicon Valley’s best-known startup accelerator, included “AI-native agencies” in its Requests for Startups, the list of ideas it wants to fund. On March 5, Julien Bek, a partner at Sequoia, published “Services: The New Software,” an essay that opens like this:
“The next trillion-dollar company is going to be a software business masquerading as a services firm”
Productizing services isn’t new. For years, productization has been used to minimize risk for a potential client, giving them a sense of predictability and transparency. But now productization has been repositioned as a way to capture the efficiencies brought by artificial intelligence. With hourly or per-project rates, that efficiency ends up in the client’s hands as a discount. With a fixed price for a defined outcome, it stays with the service provider as margin.
Postdigitalist also runs a studio that sells content services to tech companies, so this question concerns us too.
What is a productized service?
A productized service has a defined scope, a fixed price, and a standardized process. The client buys a specific outcome, not a team’s hours.
We have an example of a productized service under Postdigitalist’s umbrella: El Programa is a brand strategy sprint for founders at pivotal moments, with a predetermined process, a closed price, and a standardized outcome. We know that after X process, we’ll achieve Y impact.
But this level of productization is something of a mirage. At the end of the day, every instance of El Programa is different. Commonly, depending on the client’s needs, we decide to shorten or stretch certain modules—or complement them with other services. Despite this flexibility we’re confessing to, El Programa is far more standardized than the average productized service.
The productized-services model became particularly popular a few years ago, when it began to be adopted by small design studios and “one-person agencies.” I’d argue that while that generation of agencies popularized the concept, most of them don’t actually provide truly productized services.
What isn’t a productized service?
Many productized agencies trade a fixed fee for an “unlimited” number of deliverables—with only one workflow in progress at a time and an average turnaround of 48 hours. What predetermined outcome are they selling? That all of your design needs will be covered. But isn’t that the same thing an agency selling staff augmentation, or hour packages, is selling you?
Also, let’s assume modern working conditions, and pose a question that could well be found in an (eccentric) math textbook: if a designer delivers one asset every 48 hours (resting on weekends), and can only work on one asset at a time—how many assets can they deliver in a month?
Between 10 and 11. So why frame a service’s scope as “unlimited design” when, at most, it can deliver 10 or 11 assets to the client? What’s the bet behind a provider that obscures its scope this way? Let’s leave that there and get back to what matters.
Why now: productized services & AI-driven processes
AI lowers production costs and speeds up processes. But depending on your pricing model, this can either hurt your margins or improve them.
Let’s compare four possible pricing models and the impact AI has on each one:
| Pricing model | What the client buys | What happens when AI cuts production time |
|---|---|---|
| Hourly | The team’s time | You bill fewer hours. The efficiency passes to the client. |
| Per project, quoted by effort | A deliverable estimated in hours | In the next quote, the client expects a lower price. |
| Fixed price for a defined scope | A known package | The price doesn’t change, and the difference becomes margin. |
| By outcome | A finished unit of work | Margin grows with every process improvement, as long as the price per unit stays competitive. |
Torsten Sandor and Peter Benei, two former agency owners who co-author the newsletter AI-Ready CMO, anticipate the questions agencies that still charge the old way are going to get: why did a landing page take two weeks? Why did a ten-email campaign cost $3,000 and take four weeks? Their argument is that there’s an arbitrage window—delivering the same thing in less time while charging the traditional rate—and that window will close once other agencies start doing the same.
This connects to what we raised in our conversation about branded media: when producing becomes cheap for everyone, the differentiation shifts elsewhere. For an agency, that “elsewhere” is the judgment about what’s worth producing in the first place.
Why does Y Combinator want to invest in AI-native agencies?
Productized services and “AI-enabled services” appeared in two editions of YC’s Requests for Startups published in 2026. One detail: this category was dropped from the most recent edition, which focuses on the physical world.
In the spring edition, YC partner Aaron Epstein wrote that instead of selling clients software to do the work themselves, a company could “charge far more by using the software itself and selling the finished product at 100 times the price” (our translation, based on the text reproduced by AI-Ready CMO). YC’s conclusion is that those agencies will come to resemble software companies, with similar margins and scalability.
In the summer edition, according to the summary by VC Cafe, YC asked for “AI-native services companies” that sell the finished work rather than the tool, in insurance brokerage, accounting, tax, auditing, regulatory compliance, and health administration.
Sequoia adds a useful distinction: copilot vs. autopilot. A copilot sells the tool to a professional, who remains responsible for the outcome. An autopilot sells the finished work directly to whoever needs it. Bek gives the example of Crosby, which sells the drafting of NDAs directly to the company that needs them, not to the law firm. For an agency, the practical version of this question is: are you selling your client a new capability, or are you selling them an outcome?
The software factory Chrono Innovation is an interesting example. According to BetaKit, each of its 45 engineers works with between 15 and 20 AI agents, and the company turned its process into a product called Launchpad: the client describes the project to an agent, gets a quote within minutes, and chooses from three pricing tiers or a custom one. Its co-founder, Philippe Gratton, describes it as an attempt to move toward outcome-based pricing. The company says AI has quadrupled its output.
What should you productize?
Bek argues that for every dollar spent on software, six are spent on services—a ratio Sequoia also cites, without naming a source. His recommendation is to start with the work clients already outsource. According to Bek, if a task is already outsourced, the company has already accepted that it gets done externally, has a budget line allocated for it, and is already buying an outcome; and selling a change of vendor is always easier than selling an internal restructuring.
For a tech agency, that translates into four questions for each service line:
- Does the client already buy it externally? If an in-house team currently does it, selling it as a package means convincing them to change their organization, not just their vendor.
- Do the inputs repeat? A package works when each client provides more or less the same starting materials: a brief, access to tools, a dataset.
- Can quality be verified without being an expert? Bek notes that in standardized legal tasks, the buyer can trust the outcome without knowing the law. If your client needs your judgment to assess whether a deliverable is good, that deliverable isn’t yet a product.
- Where does the judgment live? Sequoia separates intelligence work, which follows rules, from judgment work, which requires experience. The former gets productized. The latter remains the custom layer—and is probably the part that justifies your price.
In a content agency, for example, producing pieces from an approved brief is a good candidate for packaging. Deciding what to say, to whom, and in what tone remains judgment work.
How should you charge for productized services?
There are three models in use, and each solves a different problem:
| Model | Example | What’s fixed | What stays custom | Main risk |
|---|---|---|---|---|
| Subscription | Designjoy, $4,995 per month with a discount, according to its website | Monthly price, one request at a time | The content of each request | Clients who fill up the queue and strain capacity; a blurry line between staff augmentation & real productization |
| Fixed-scope package | Launchpad’s tiers, El Programa | Scope and price of each tier | Specific scope | Requests that don’t fit any tier |
| By outcome | Sierra, $5 per support ticket resolved | Price per finished unit | Nothing, in theory | Defining what counts as an outcome and who verifies it |
The Sierra example is explained well in an interview with Bek summarized by TBPN Digest: the company charges $5 per resolved ticket, compared to roughly $20 it would cost to resolve the same ticket with a human agent, according to that comparison. Sierra isn’t an agency, but it shows the model in its purest form.
Warning signs
Productizing badly is more expensive than not productizing at all. Let’s look at the most common failure modes:
Commoditization
If every agency uses the same AI models for the same package, prices converge downward. The package needs something that doesn’t come from the model: proprietary data, vertical specialization, or judgment.
Promising more autonomy than you actually have
In the vast majority of cases, no matter how refined the prompt, an AI’s output will range from mediocre to worse. If you provide a complex service and want to preserve its quality and your brand’s standing, talented humans will have to step in on whatever the AI produces. If you’re selling an “AI-powered” outcome that actually depends on a team’s hours, you’re presenting as hyper-scalable something you should really be selling as handcrafted.
The boring explanation
There are very good reasons to view the promise of “AI-native”/”AI-enabled” productized services with skepticism. Let’s start from the premise that having a bunch of VCs (however famous) line up behind an idea doesn’t mean the idea works. Investors need new categories to fund, and “services with software margins” makes for a good story in a funding round.
HFS Research published a report on agentic services, in which it evaluated 36 service providers using data from September 2025 to January 2026. It concludes that the model it calls Services-as-Software “remains limited,” with autonomy constrained by human oversight and accountability for outcomes. Only five large firms (Accenture, Cognizant, IBM, Infosys, and Wipro) are designated by HFS as the most advanced, and the greatest autonomy is seen in narrow areas like IT service management and software development.
The skeptical reading explains well why investors talk about this topic so much. It doesn’t explain the pricing mechanism, which works even when autonomy is partial: an agency that cuts, say, 30% off its production time gains margin under fixed pricing and loses it under hourly billing, with no autonomous agent required. The evidence favors a more modest version of the thesis: you don’t need to become a software company, but you do need to stop charging for an input that AI is making cheaper.
What would change our minds?
This piece assumes AI keeps lowering production costs faster than clients adjust what they’re willing to pay. If, over the next twelve months, buyers of services start demanding package prices calculated off the cost of the models rather than the value of the outcome, the margin available from productizing shrinks, and the advantage goes back to being purely about judgment. See you in October 2027.
By Aaron Marco Arias