Eupraxic

Articles

Mastra now rents out its engineers. What that tells you if you run a 20-person SaaS

Mastra's new FDE program is built for Salesforce, MongoDB, and Workday. Two questions tell a smaller company whether it fits, and what to buy if it doesn't.

By Nasser Ghanemzadeh · · 5 min read

I'm on Mastra's newsletter list, and on November 10 its CEO, Sam Bhagwat, emailed the list with a new offer. Mastra makes an open-source TypeScript framework for building AI agents. This time the offer was people: you can now bring engineers from Mastra's own team into your codebase, for what the company's post calls "less than the cost of one senior AI hire."

He made the same announcement on X. After saying that Mastra had already helped teams at its largest customers ship agents, he wrote:

So today we're launching this as a program: Mastra FDE.

It was Mastra's third services launch in five weeks. An agency program for consultancies that build on the framework arrived on October 8, and a custom harness program followed on November 5. The engineers themselves came five days after that.

A framework company has every commercial reason to tell you its product is easy to adopt. This one told its whole mailing list that the fastest route to a production agent is a person who has shipped one before, working next to your team.

I think that's right. I also think it matters most to companies the program wasn't designed for.

This is bigger than one framework

Mastra is following the largest companies in the field. In early May, Anthropic announced a $1.5 billion enterprise deployment venture backed by Blackstone, Hellman & Friedman, and Goldman Sachs. Days later, OpenAI launched the OpenAI Deployment Company with more than $4 billion in committed capital and agreed to buy Tomoro, a UK consultancy, which brings it roughly 150 engineers and deployment specialists (Decrypt).

So the model makers got there in the spring and a framework maker got there in the autumn. Each of them sells software, and each has concluded that customers still need a person to get the first production system working.

The job title draws some eye-rolling, and its history explains why. Tom Hollands of Andreessen Horowitz traced it to 2011, when Palantir gave its solutions and integration engineers a new title. He calls the move "title arbitrage" and separates it from plain title inflation with one test: does the new title describe work that nobody would have recognized five years earlier?

For AI deployment, I think the title passes. Writing evals for an LLM workflow and wiring it into a product that paying customers use was nobody's job in 2021.

Who these programs are built for

Look at the customers named in Mastra's announcement: Salesforce, MongoDB, and Workday. Salesforce built the harness behind Agentforce Vibes on the framework, and about 100,000 developers use it by default. Workday moved its security operations agents from Bedrock Agents to Mastra in 31 days. These companies have platform teams, so the engineer Mastra sends has colleagues to pair with on the other side.

The engagement is shaped for that buyer. Mastra offers two modes: 6 hours a week of advice, or 20 to 40 hours a week of building. Hours and deliverables are scoped case by case, the package includes Mastra's enterprise product, and no price is published.

I'd sell the same program if I ran a framework company. It fits the customers who are already paying.

Now read it as the founder of a 20-person B2B SaaS company. You have six engineers and a roadmap that's already full, and nobody on the team has put an LLM into production. What you need is one workflow running in your product by a date you can plan around, at a price you know in advance. A block of weekly hours from a framework specialist, with no price on the page, is a bigger purchase than that.

Your three options

There are more than three, but these are the ones Mastra's launch puts in front of you.

Hire an AI engineer. This is the default, and Mastra's own post names the problem with it: hiring takes months, needs some blind trust in the candidate, and is costly. My numbers agree. The 74 US forward deployed engineering roles I found in October that post a salary range had a median midpoint of $216,104 in base pay. Hiring makes sense once AI is a permanent line on your roadmap. It's a slow way to find out whether your first workflow was worth building.

Rent the vendor's engineer. You get someone who knows the framework from the inside and has seen it run in production at other companies. The cost is that you choose the framework before you've fully defined the problem, because a vendor's engineer will build on the vendor's product. That's a fair trade when your stack is already decided. When it isn't, you've committed early, and a first deployment often needs less framework than you'd expect.

Bring in an independent forward deployed engineer. Disclosure: this is what I do. An independent works in your existing stack and has no license to sell you, so the tools get picked for the problem. The weakness is that one person has no product team or roadmap behind them. Ask for a fixed scope in writing with a refund clause, and for a handoff your own engineers can run without the person who built it.

A fair test

Two questions tell you whether a vendor program fits.

  1. Have you already chosen your agent framework? If your team builds on Mastra today, their engineers will be faster than anyone else, me included.
  2. Do you have engineers who can pair with an embedded specialist for 20 hours a week? If you don't, a lot of what you pay for is advice nobody has time to absorb.

Two yeses point you to the vendor. Take the call with Mastra.

If either answer is no, ask a different question: can you name the one workflow you would ship first? If you can, you have a scoped project, and a scoped project can have a fixed price and an end date. If you can't, a scoping conversation is the right first purchase, before anyone's hours.

What I do with ten days

I run the small version of this. The FDE Sprint is ten business days and one AI workflow, built inside your product on your stack, at a fixed scope and a fixed price: $7,500 fixed. Day 1 ends with a signed one-sentence Brief. If there's no working prototype on your real data by Day 5, or the workflow isn't running by Day 10, you get a full refund.

At the end your team has the code and the rubric it was scored against, along with a playbook for running the workflow without me. I call the result your first autopilot: one workflow that does the work, where a copilot would only help a person do it.

Mastra's announcement is good news for this model. The people who build the frameworks now say in public that their customers need an engineer alongside the software, and that holds for a 20-person company too, at a smaller size.

If you have a workflow in mind, book a 15-minute scoping call. I'll tell you whether it fits in ten days. If a vendor's program would serve you better, I'll tell you that too.

About the author

Nasser Ghanemzadeh is a former CEO and CPO with 17 years of product leadership. He exited Nivo, a B2B fintech with 200K+ users, after growing revenue 450%, was founding head of product at Pangouan, and founded the Finnova accelerator, which he ran as its first managing director. He builds Vectig with Claude Code and MCP every day and runs every Eupraxic Sprint himself. Eupraxic is one engineer for now; the method is written down so it will not stay that way.

This article draws on his book, Forward Deployed AI Engineering.

Want one useful AI workflow in your product?

Score your workflow in two minutes, then book a 15-minute Scoping Call. If it fits, you leave the call with a Brief draft and a kickoff date.