Why Product Creation Needs Agents, Not More SaaS

I want you to think about the last product your company launched.

Not the marketing. Not the sales figures. The actual process of getting it from an idea in someone's head to a physical thing a customer could buy. Think about how many people touched it. How many tools were involved. How many emails were sent. How many versions of the same file existed in how many different formats on how many different platforms.

If you're in fashion, footwear, or consumer goods, I already know the answer. The process took somewhere between 10 and 16 weeks. It involved a minimum of four separate teams who barely talk to each other. It required at least six different software tools that definitely don't talk to each other. And somewhere along the way, someone re-entered data by hand into a spreadsheet because the system that had the information couldn't export it in a format the other system could read.

I know this because I've lived it for nine years. And I know it because every single enterprise team I've spoken to in the past two years tells me the same story with different names attached.

The question nobody seems to ask is: why?

The six-tool trap

Here's what a typical product development workflow looks like at a mid-size brand in 2026.

The creative director opens Miro to build a mood board. Maybe Illustrator for initial sketches. They arrive at a concept. That concept exists as a collection of images and notes on a digital whiteboard that nobody outside the design team has access to — or frankly, understands.

The technical designer takes the concept and rebuilds it in CLO3D for 3D rendering, or more commonly, in a combination of Illustrator and Excel. They create a tech pack — the single most important document in physical product creation — by manually typing measurements, materials, and construction notes into rows and columns. This tech pack doesn't reference the mood board. It doesn't automatically inherit anything from the design concept. It's a re-creation from scratch, interpreted through the technical designer's understanding of what the creative director intended.

The sourcing agent takes the tech pack and emails it to three or four factories, asking for costing and samples. The factories respond with pricing in different formats, using different assumptions about materials, minimum order quantities, and production timelines. The sourcing agent builds a comparison spreadsheet. The information lives in their inbox and on their laptop.

Meanwhile, the marketing team is waiting. They can't create content until they have product photography. They can't get product photography until physical samples exist. The samples take two to four weeks to produce and ship internationally. When they arrive, they often don't match the original design intent because something was lost in translation between the concept, the tech pack, and the factory's interpretation.

So the process loops. Revision notes go back. New samples are requested. Another two to four weeks.

When the final product is approved, the marketing team books a photography studio, hires models, and shoots every SKU. The photos are edited in Photoshop and Lightroom. Campaign assets are built in Canva or InDesign. Social media content is created separately.

I'm not describing a broken company. I'm describing the industry standard. This is how the best-run brands in the world still operate.

Six tools. Four teams. Zero integration. Fourteen weeks.

Why more software doesn't fix this

Over the past decade, the fashion industry has adopted a lot of software. PLM systems like Centric and Backbone to track the product lifecycle. 3D tools like CLO3D and Browzwear for garment simulation. Photography tools like Photoroom and Flair for AI-generated product images. Each of these tools is excellent at what it does.

The problem is that they're islands.

Your PLM system doesn't know what your designer just created in Illustrator. Your 3D tool doesn't generate a tech pack. Your photography tool doesn't know the material specs of the product it's shooting. Your costing is done in Excel, disconnected from everything.

When you add a seventh tool to a stack of six, you don't reduce fragmentation. You increase it. You now have seven places where data lives, seven formats to reconcile, seven systems that need to be manually kept in sync by humans. The coordination cost — the hours spent ensuring that the information in Tool A matches the information in Tool B — grows with every addition.

This is the SaaS trap. Each tool optimizes one step while making the overall system more complex. McKinsey's State of Fashion 2026 report captured this problem precisely: up to 90% of AI initiatives in fashion fail to scale beyond the pilot phase, predominantly because the underlying technology and data infrastructure is fragmented and insufficient.

The tools aren't the problem. The architecture is.

The agent model

There's a reason software engineering went through a similar transformation. Five years ago, developers used separate tools for writing code, testing code, deploying code, and documenting code. Each tool was good at its step. Each tool was siloed.

Then GitHub Copilot appeared. Not as a better code editor, but as an agent that understood the developer's intent and could participate across the workflow. It didn't replace the developer's judgment. It replaced the mechanical work of translating intent into implementation.

The shift was architectural. Instead of a human orchestrating a collection of tools, an intelligent agent worked alongside the human — remembering context, understanding the project's history, and handling the repetitive work that consumed most of the developer's day.

Product creation is waiting for the same shift.

What if, instead of six tools that a human manually coordinates, you had agents that understand your brand and collaborate across the entire lifecycle?

An agent that takes your sketch and turns it into a photorealistic render — and then, because it understands the product it just designed, generates a manufacturing-ready tech pack with accurate measurements, materials, and construction notes. An agent that calculates production costs across multiple regions — China, Vietnam, India, Italy — because it knows the exact specifications from the design step. An agent that creates on-model photography and campaign content from the same product data, maintaining color accuracy and brand consistency because it shares context with the agent that designed the product.

One conversation. One intelligence. Every step connected.

This isn't a fantasy. This is what we've built at IMAI.

What makes agents different from tools

The difference between an agent and a tool isn't just a branding exercise. It's an architectural distinction that matters for three specific reasons.

Agents have memory

A tool is stateless — you configure it the same way every time. You open Illustrator, and it has no idea what you designed yesterday. You open your PLM, and it doesn't learn from your past decisions.

An agent remembers. It knows your brand's Pantone standards. It knows which fabrics you've used in past seasons. It knows that you rejected the coral colorway last time and preferred the sage. It knows your factory in Tiruppur has a minimum order of 500 pieces and a lead time of six weeks. Every interaction makes it more aligned with how your brand works. The fiftieth product you create is dramatically better than the first — because the agent has learned.

Agents collaborate across functions

Tools are siloed by design — each one does its thing in its own interface. An agent can reason across domains. When our Design Agent creates a product, our Manufacturing Agent can immediately generate the tech pack because they share a unified understanding of the product. The photography our Marketing Agent produces is accurate to the design because it reads from the same structured product representation. There's no handoff. No file conversion. No "let me export this and upload it to the other system."

Agents reduce coordination, not just execution time

Most AI tools promise to make individual tasks faster — generate an image in seconds instead of hours, create a render in minutes instead of days. That's valuable, but it's not transformative. The real cost in product creation isn't the execution time of any single step. It's the coordination time between steps. The emails. The revision cycles. The meetings to align teams. The weeks spent waiting for someone else to finish their part.

Agents eliminate coordination because they hold the full context. There's nothing to coordinate when the same intelligence handles design, specs, and marketing.

This is bigger than fashion

I've described the fashion industry because that's where we started. But the problem is universal across physical product creation.

Furniture brands design in CAD, prototype with physical samples, photograph in studios, and manage specs in spreadsheets — the same fragmented workflow. Beauty brands concept packaging in Illustrator, build specifications separately, and create marketing content independently. Home goods companies, consumer electronics brands, toy manufacturers — every industry that makes physical products suffers from the same architectural problem.

The $2.24 trillion global market for clothing, footwear, and accessories alone is just the beginning. The total addressable market for AI-powered product creation spans every physical consumer good category.

The timing

Why now? Why couldn't this have been built five years ago?

Because the AI models weren't capable. Generating a photorealistic product render requires multimodal understanding — interpreting a sketch, reasoning about fabric physics, applying brand color standards, and producing an image that's accurate enough for buyer approval. Generating a tech pack requires structured reasoning about measurements, materials, and construction methods. Connecting these outputs into a coherent product representation requires cross-domain reasoning that simply wasn't possible before the current generation of foundation models.

The models are now good enough. What's needed is the domain expertise, the data pipelines, and the architectural decisions to build specialized agents on top of them. That's the work.

The endgame

I want to describe where this goes, because the trajectory matters more than the current state.

Today, IMAI's agents serve individual brands and manufacturers. A brand uses IMAI to design a product, generate specs, and create marketing content. A manufacturer uses IMAI to respond to buyer briefs with tech packs and costing.

Tomorrow, when a brand designs a product on IMAI and a manufacturer responds to that brief on IMAI, they're working on the same platform. The brand's design intent flows directly to the factory's production capability. There's no handoff. No file transfer. No miscommunication.

That's a network effect. And it's a fundamentally different business from selling SaaS seats.

The companies that define the next decade of physical product creation won't be the ones that build better tools. They'll be the ones that build the infrastructure where products are designed, specced, and launched — where both sides of the table meet.

We're building that infrastructure. And we're just getting started.