Mechanical Orchard updates mainframe migration with AI

Two-year-old startup Mechanical Orchard migrates mainframe apps to the public cloud using generative AI technology.

Mainframe systems are used by 44 out of the top 50 banks, ten out of ten top insurers, and 23 out of 25 top US retailers. Some 71 percent of the Fortune 500 are still reliant on mainframes, and spend in excess of $20 billion a year buying and maintaining them. The mainframe is a restricted and costly environment. Moving apps from it to the public cloud means they can interoperate easily with open source code, get away from historic COBOL programming, become SaaS-style entities, and be more agile and able to develop their previously fossilized code.

Rob Mee, Mechanical Orchard
Rob Mee

Mechanical Orchard was founded in San Francisco in 2022 by three ex-Pivotal Labs guys, CEO Rob Mee, COO Matthew Work, and VP R&D Dan Podsedly. It took $7 million in a seed round that year and has just raised $24 million in an A-round, having demonstrated substantial progress. It has 50 staff and offices in the UK, Ireland, Italy, and Germany. Customers include Omni Logistics.

‍It says it uses iterative, AI-enhanced, and reverse engineering approaches to move large enterprise apps off mainframe systems and into the cloud. During a briefing in London, Mee told us that Mechanical Orchard’s method was to look at a mainframe or other legacy system application and migrate it to the public cloud incrementally, module by module. Mee said “there is no silver bullet, and never will be” that can be used to move legacy apps and their components en masse to the public cloud. This migration requires a meticulous, iterative, and incremental approach.

Mainframes have distinctive hardware and system software architectures that do not translate easily to x86 compute, commodity-based storage, and networking-based infrastructure. There is no straight equivalence between mainframe Direct Access Storage Devices (DASD) and the main SSD and HDD instances available in AWS, Azure, and Google. Mainframe apps could also have code components that are decades old and may not have documented source code.

A typical customer engagement starts with a pilot and looks like this. It selects a piece or component of the application that has defined inputs and outputs. If it doesn’t have the source code, Mechanical Orchard treats it as a black box with data flows, which it reverse engineers into a cloud-native application. This is loaded into an isolated sandbox and tested with various inputs, amended as needed when the output is different from what is expected, until it parallel runs with the same outputs as the source legacy module and is just as fast if not faster. This pilot can take six to eight weeks. 

That is then subject to the customer’s judgement as to whether to take it further. If it’s a yes, component by component, the mainframe application is reverse engineered to use cloud-native code based on public cloud compute, networking, and storage instances. It then runs in parallel with the source mainframe component with equivalence monitoring looking for and detecting different outputs that need understanding and correcting, and checking performance.

Generative AI is being used to help build a visual representation of the mainframe app component and what it does. Mee said AI can synthesize data, identify coding patterns, and speed up code development and documentation. It helps Mechanical Orchard staff be more productive.

When full equivalence is attained, says the firm, the customer can switch over from the mainframe to the public cloud component. The switchover point varies with the criticality of the software. A financial trading module will be tested far more severely than a retail system for example. AWS, Azure, and Google’s public clouds are supported. In fact, the three partner with Mechanical Orchard because it represents a mainframe app and data on-ramp for them. Mechanical Orchard represents a different approach from using public cloud mainframe emulation, which leaves the source app still frozen.

This is a service-based operation such as Arthur Andersen or Price Waterhouse might provide. For customers, they see that the Mechanical Orchard process costs money, but when it is complete they have a migrated mainframe app that gives them the ability to reduce mainframe costs and move into a far more scalable, agile, and fast-moving development environment than mainframe land. Mainframe digital transformation moves at glacial speed. Cloud-native development – DevOps with continuous integration and delivery – is much faster and there is a far larger base of skilled coders available. 

That’s Mechanical Orchard’s would-be USP – dependable mainframe jailbreaking that is carried out proof point by proof point, component by component, until the frozen mainframe app is public cloud resident and ready to be developed as the business needs.


Pivotal Labs, an application development, containerization, and Kubernetes business, was founded in 1989. It was acquired by EMC in 2012 and spun-out in 2013. Rob Mee founded Pivotal and became the CEO that year. The company went public in April 2018. VMware completed a $2.7 billion acquisition of Pivotal in December 2019, at which point Mee left.