Fastly AI grew out of a simple problem I kept seeing: useful image editing tasks were still split between complicated professional software, specialist services, and separate single-purpose websites. I wanted to make those workflows accessible in one browser-based product without requiring users to become professional editors.

Today, Fastly AI provides six focused workflows: background removal, 4K image enhancement, AI product photography, image outpainting, ID photo creation, and photo restoration.

What makes us different is our focus on practical outcomes rather than a generic AI interface. A customer uploads an image, chooses the relevant options, sees the credit cost, and downloads a finished file. We try to make every workflow simple, predictable, and easy to evaluate.

Most of our new business currently comes from organic search and practical how-to content. People usually find us while searching for a specific outcome, such as removing a background, improving a low-resolution image, restoring an old photograph, or adapting an image to a different aspect ratio.

That traffic is valuable because the customer already has a real task to complete. Over the next year, I would like more growth to come from customer recommendations, trusted product comparisons, and AI assistants that can match a user’s problem with the right workflow.

I would rather earn repeat discovery through useful tools and reliable results than depend heavily on short-term paid traffic.

In the last year, I have handed much of the first-pass software development work to AI agents. They help explore the codebase, trace dependencies, research implementation options, draft changes across multiple files, and create an initial prototype. This removes repetitive research and lets us test ideas faster.

What I moved back to the human side is final product judgment. I tried giving AI more influence over what should ship and how final customer-facing content should read, but it can complete the immediate task while missing business context, edge cases, or brand judgment.

AI can research, propose, and produce a first pass. A person still owns architecture, security, testing, factual accuracy, product priorities, and the final release decision.

Customers increasingly arrive with more specific vocabulary and a clearer idea of the result they want. Instead of asking whether AI can improve an image, they may ask for a transparent PNG, a 4K upscale, a 9:16 expansion, or a product image suitable for an online store.

Search results, comparison content, and AI assistants are helping people understand what is possible before they reach a product. We are starting to see AI assistants become part of the discovery process, but I would not describe ChatGPT as a primary acquisition channel yet without consistently attributable traffic.

The larger change is that customers now expect a direct answer, a simple workflow, and a usable result much faster.

For the next 12 months, I would tell other AI software founders to select two or three workflows where the input, output, owner, and quality check are clearly defined. Measure the current cycle time, rework, cost, and customer outcome before introducing AI, then measure the difference.

Keep human accountability for security, factual claims, architecture, and customer-facing decisions that could cause harm. AI should accelerate responsible work, not remove responsibility.

I would ignore the pressure to add a chatbot to every screen, automate every decision, or chase every new model release. Access to models is becoming easier. The more durable advantage is the workflow around the model: product context, evaluation, distribution, customer experience, and trust.