• Failory
  • Posts
  • The Vegetable Middlemen

The Vegetable Middlemen

Otipy’s human-powered network couldn’t scale fast enough

Hey — It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 5 most important things:

Let’s get into it.

This Week In Startups

🔗 Resources

How Tech workers really feel about work right now.

Vertical AI's Integration Problem.

📰 News

Odyssey’s new AI model streams 3D interactive worlds.

AI may already be shrinking entry-level jobs in tech, new research suggests.

💸 Fundraising

Identity security automation startup Cerby raises $40M.

Elon Musk’s Neuralink raises $600M at $9B valuation.

Fail(St)ory

The Fresh Chain

This week, Otipy, a grocery delivery startup that once promised to connect farmers to urban homes, shut down.

Backed by $44 million and riding the tailwinds of pandemic-era demand, the company quietly disappeared after failing to raise another $10 million. Hundreds of employees and gig workers were left hanging.

What Was Otipy:

Otipy launched in 2020 with a mission to deliver fresh produce from farms directly to consumers’ homes. But instead of doing doorstep delivery themselves, they leaned on a network of local resellers — mostly homemakers, shopkeepers, and small-time entrepreneurs — who handled last-mile logistics.

Here’s how it worked: customers placed orders on the app. Those orders went to nearby resellers, who would batch them and deliver within their community the next morning. Resellers got a cut of each sale, and Otipy got to scale delivery without building a huge fleet.

The operational flow was efficient: produce was sourced directly from farmers, transported to Otipy’s warehouses for quality checks, and then dispatched to resellers for final delivery. By cutting out multiple intermediaries, Otipy aimed to reduce costs and ensure fresher produce for consumers.

To manage inventory and minimize waste, Otipy employed predictive algorithms that analyzed customer ordering patterns. This data-driven approach allowed them to anticipate demand and adjust procurement accordingly, reducing the typical 30-40% waste seen in traditional supply chains to around 3%.

While the model had its merits, like supporting local economies and promoting fresher produce, it wasn’t without challenges. Coordinating a vast network of resellers required significant logistical oversight, and as consumer preferences shifted towards instant delivery models, Otipy’s scheduled delivery system faced increasing pressure.

By 2024, the startup was still alive and pulling decent revenue — around $20 million for the year, up from roughly $14 million the year before. But margins were thin, competition was heating up, and Otipy was burning through cash. Behind the scenes, they were already looking for another $10M to stay afloat.

The Numbers:

  • 🏁 Founded: June 2020

  • 💰 Funding: $44M (equity + debt)

  • 🧍‍♂️ Team: ~300 employees + gig workers

  • 📈 FY24 revenue: $20 million USD

Reasons for Failure: 

  • The fundraising never came: Otipy needed another $10 million to stay alive. The round was supposed to extend their Series B. But it never closed — and without it, the company had no buffer. One day, it just stopped. Former employees say they were told the money was coming. Then everything went quiet. There was no real announcement. Just radio silence, missed salaries, and wallets that still show positive balances.

  •  Quick commerce changed the game: When Otipy launched, the idea of pre-ordering fresh produce and getting it next morning felt efficient. But by 2023, speed became everything. Blinkit and Zepto trained users to expect 10-minute delivery. Suddenly, “fresh tomorrow” looked like “late.” Otipy, with less funding and a heavier model, couldn’t keep up.

  • The model was smart, but heavy: The reseller network gave Otipy an edge in building trust and reducing delivery costs. But scaling it meant dealing with hundreds of semi-independent agents — each with their own systems, expectations, and issues. 

  • Founders might’ve waited too long to pivot: They raised a lot, built a team, got to product-market fit, but maybe they stuck to the original model too long. Quick commerce didn’t appear overnight. If they had shifted earlier — changed the model, narrowed the offering, cut burn — the story might be different.

Why It Matters: 

  • Fundraising is not a strategy. If your survival depends on a wire transfer, you’re already out of time.

  • Markets shift. Otipy launched into a world of morning deliveries. It died in a world of 10-minute commerce.

  • Heavy operational models are risky. Great when they work — brutal when they stop.

Trend

Microsoft Build 2025

Last week, I talked about Google I/O and all the wild stuff they dropped — like those AI-generated Veo 3 videos that looked straight out of a sci-fi movie. But while Google was showing off to creators and casual users, Microsoft was busy talking to a different crowd.

At Build 2025, they laid out a vision that’s less flashy but potentially more important: an internet where AI agents actually do work. Not just answer questions or summarize stuff, but take action — fix bugs, run experiments, manage your internal tools.

Microsoft calls it the “agentic web.” It’s their big bet on how the internet (and work) is going to change.

Why It Matters:

  • Microsoft is building for businesses, not browsers. Their tools aren’t meant to impress — they’re meant to plug into your workflow.

  • AI agents are becoming coworkers. The shift from assistant to operator is happening fast.

  • If you’re building anything — especially SaaS — this is the infrastructure you’ll be using (or competing with) soon.

Key Announcements

GitHub Copilot Becomes a Real Teammate

Until now, GitHub Copilot has been a decent autocomplete. Now, it’s more like a junior dev. You assign it an issue — a bug, a feature, whatever — and it spins up a secure environment, checks the codebase, writes the fix, and opens a draft pull request. It even pulls context from old issues and comments.

It’s not perfect. But if you’re running a small team or maintaining a legacy codebase, offloading even 20% of your tickets is a huge win.

Azure AI Foundry Is Now a Giant Toolbox

Azure Foundry is turning into a one-stop shop for building AI agents.

It supports over 10,000 models now, including open-source and niche task-specific ones. You can fine-tune them, hook them into your tools, and deploy agents that don’t feel like generic chatbots.

It’s also integrating deeper into Microsoft’s stack. So if you’re already on Azure or use Copilot Studio, this is the path of least resistance.

Copilot Learns Your Company’s Voice

You can now train Microsoft 365 Copilot to sound like your company. It learns your tone, your workflows, your docs. And you don’t need to write a single line of code. This matters more than it sounds — internal AI tools are only useful if they feel native to your org. Otherwise, people stop using them.

NLWeb: AI Chat for Every Site

NLWeb is like HTML but for adding chatbots. It lets you drop a natural language interface into any website with a few lines of code. Pick your model, connect your data, and that’s it — users can now “talk” to your content.

It’s simple, and it’s open-source. If you’re running a product where users need support, onboarding, or explanations, this is worth looking at.

Discovery: AI Agents for R&D

Microsoft also introduced Discovery — a research tool that uses AI agents to simulate experiments, form hypotheses, and crunch data. It’s built for scientists, but the big idea here is that AI isn’t just summarizing results anymore. It’s generating them.

To show it off, Microsoft claims Discovery helped them invent a new data center coolant in under 200 hours. That’s the kind of thing that usually takes a year.

Microsoft Build vs Google I/O

The contrast between the two conferences couldn’t be clearer.

Google is building AI for consumers. Their announcements were all about helping individuals search better, shop smarter, and create content faster. The vibe was: “Here’s a smarter assistant for your life.”

Microsoft is building AI for enterprises. Their focus was tools for teams, infra for developers, and agents that plug into company systems. The vibe was: “Here’s a junior employee who shows up to work and doesn’t complain.”

This split matters. If you’re building B2C, Google’s direction tells you what your users will expect. If you’re building B2B, Microsoft is shaping the tools your clients will adopt — or expect you to integrate with.

Help Me Improve Failory

How Was Today's Newsletter?

If this issue was a startup, how would you rate it?

Login or Subscribe to participate in polls.

That's all of this edition.

Cheers,

Nico