Listen to this blog as a podcast:

Estimated time to read: 6 minutes

I’m Leonardo, Engineering Manager at Algolia, where I lead our Generative AI team. A few weeks ago I gave this presentation — From Months to Minutes: Building Production AI Agents — at our annual community conference, DevCon. This is a summary of that presentation. If you’re interested in this, you may want to check out all the great talks.

For the past three years, I’ve been obsessed with one question:

How do we take a generative AI experience from a fun prototype to a real, production-ready product in the hands of users?

Here’s the honest truth: prototyping AI is fast—ridiculously fast. You can spin up a demo in a day, maybe even an afternoon. But taking that demo to production? That’s where things get painful.

The walls we hit (over and over again)

1. The infrastructure wall

Our first big prototype we built at Algolia was a shopping assistant. In a single day, we hacked it together—hooked up a model, pulled data, and watched it produce magical results. It looked so good we thought, this is ready to ship.

Then our ops team asked the real questions:

Can you deploy it reliably?
What happens when 1,000 users hit it at once?
How’s your latency? What’s your failover plan?

Suddenly our one-day demo didn’t look so ready anymore. Because it’s not just about a model that answers questions—it’s about everything around it: infrastructure, scaling, reliability. Building something that works once is easy. Building something that works every time, for everyone—that’s production.

2. The compliance wall

Just as we made progress, our legal and compliance teams stepped in—rightfully so.

“What about GDPR? Do you have audit logs? Where’s the data stored? Who can access it?”

None of that existed in our prototype. No audit logs. No policies. Just a shiny demo. Weeks disappeared into privacy reviews, security questionnaires, and data protection assessments. Compliance is essential—but it’s never accounted for in the prototype phase, and that gap slows every team down.

3. The engineering wall

Then our own developers pushed back:

“We need authentication. We need logging. We need retries, monitoring, guardrails, dashboards.”

They were absolutely right. Every real product needs those foundations. Yet every new AI project meant rewriting the same glue code over and over again—rebuilding authentication, logging, and monitoring from scratch. It wasn’t innovation; it was repetition.

4. The product wall

Finally, product management joined the conversation:

“What metrics are we tracking? How do we measure user happiness? How will we improve this over time?”

Those are the right questions. But again, none of that existed in our demo. No feedback loops. No dashboards. No user signal tracking. That’s when it hit us: what we thought was almost done was actually just the beginning.

Is this starting to sound familiar with some of your own agentic prototypes? 

The turning point

After hitting the same walls project after project, we finally asked ourselves a hard question:

Why are we doing this to ourselves?

Every new AI project started the same way. Every new agent rebuilt the same foundations. Even with Algolia’s years of infrastructure and search experience, we were still stuck in the loop.

We had two choices:

1. Keep rebuilding the same foundations forever.

2. Solve the problem properly—once—and make the solution available to everyone.

Naturally, we chose option two. 😀

Introducing Agent Studio

That’s why we built Agent Studio—Algolia’s platform for creating generative AI experiences that are simple, fast, and production-ready.

Everything we wished we had when we were trapped behind those walls—hosting, compliance, boilerplate, monitoring—it’s all built in. Instead of reinventing the wheel every time, you can just build.

Let me show you how.

Rebuilding the shopping assistant (the right way)

Inside the Algolia dashboard, under Generative AI Experiences, you’ll now find Agent Studio (beta). Here’s how easy it is to stand up a real, production-grade AI agent.

Step 1: Create an agent

 

You can start from scratch or choose a template—say, our “Shopping Assistant.”  Each agent has an Instructions section where you define how it behaves: tone of voice, number of results, fallback behavior, or how to handle tricky cases (like competitor queries). You’re not just tweaking prompts—you’re defining the agent’s personality and guardrails.

Step 2: Connect to your data

With the built-in Algolia Search Tool, you can connect your agent to your existing indices—products, movies, whatever powers your business. Just select the index, describe it, and you’re done. No boilerplate. No custom backend. Your agent can now access your data instantly.

Step 3: Choose a model provider

Next, pick your model. Some models are faster and cheaper; others excel at reasoning or complex instructions. You can easily swap providers and see how your agent’s responses change—you’re always in control.

Step 4: Publish and Integrate

Once you’re happy, hit Publish. You get a ready-to-use API endpoint and sample frontend code (using the Vercel AI SDK).  In minutes, your agent can be live in any app or website.

No backend development. No ops setup. Just plug in your agent ID and start testing real use cases.

From Prototype to Production in Minutes

To prove it, I asked my rebuilt shopping assistant:

“I’m running a marathon next month. Can you suggest the gear I’ll need?”

It instantly recommended running shoes, technical socks, and apparel—all drawn from live product data. And here’s the best part: this isn’t a fragile demo. This is production-ready—with hosting, compliance, monitoring, and scaling already handled.

Turning Walls into Foundations

Those “walls” that used to slow us down—hosting, compliance, boilerplate, monitoring—are now foundations. With Agent Studio, we’ve compressed months of engineering overhead into minutes of setup.

Now, instead of getting stuck in the loop of rebuilds, you can focus on what actually matters: innovation.

TL;DR: What Agent Studio Delivers

• Production-grade hosting, monitoring, and compliance
• Instant data connection with Algolia indices
• Model flexibility and experimentation
• Built-in feedback and observability
• Real products—not prototypes

From months to minutes.

That’s what happens when you stop treating AI as a side experiment and start building it on real foundations. You can try Agent Studio for yourself. If you don’t already have an Algolia account, sign up for a free account. Need more info? Contact us and we can provide a demo or answer any questions you might have.