I love a good pivot story.
This is how Bloop pivoted to find a problem that companies are spending $100s-of-millions to solve.
Bloop - YC and early customers
Bloop was founded in January 2021 by Louis Knight-Webb and Gabe Gordon-Hall and were quickly accepted into YC.
Their initial pitch in March 2021 was "AI search over all of the code in GitHub", which had a brief moment of relevancy before GitHub Copilot launched in October 2021.
By 2023, they had pivoted to "Conversational search over any private repo" but found that most users preferred to use other products like Cursor and GitHub Copilot which had bundled much of the same functionality.
The Bloop team had predicted two massive trends, made two shots on goal, but neither had landed.
Bloop's landing page in February 2023A possible pivot
Although the Bloop team didn't capture the zeitgeist like Cursor and Copilot, by 2024 they had real users and importantly, real customers paying them real money to talk to their codebases.
But Louis shared how one conversation with a huge company made them rethink Bloop:
"We got like a 1,000,000,000 lines of COBOL in production, can it do that?"
But COBOL was completely new to Louis:
I was like: what's what's this COBOL thing? I've kinda heard of it, and everybody's got an older relative who has worked on it or something like that. But it was it was it was a completely new environment to us: the mainframe.
This was a very hard pivot: killing a good opportunity in search of a great opportunity.
What is COBOL?
"Common business-oriented language (COBOL is a high-level, English-like, compiled programming language that is developed specifically for business data processing needs"
COBOL has a hierarchical structure that comprises divisions, sections, paragraphs, sentences, verbs and character strings.
Here is a hello world example in COBOL
IDENTIFICATION DIVISION.
PROGRAM-ID. IDSAMPLE.
ENVIRONMENT DIVISION.
PROCEDURE DIVISION.
DISPLAY 'HELLO WORLD'.
STOP RUN.
COBOL is still heavily used, especially in financial services:
- COBOL’s imperative, procedural and (in its newer iterations) object-oriented configuration serves as the foundation for more than 40 percent of all online banking systems
- It supports 80 percent of in-person credit card transactions
- It handles 95 percent of all ATM transactions
- It powers systems that generate more than USD 3 billion of commerce each day.
What is a mainframe?
"Mainframes are data servers that are designed to process up to 1 trillion web transactions daily with the highest levels of security and reliability."
And importantly, they are still very much in use:
"Since the advent of the internet and the rise of cloud computing, some may think of the mainframe as a tech dinosaur. On the contrary, the mainframe evolved to keep pace with other technologies and continues to play a vital role in IT infrastructure."
In a 2022 report, IBM found: "45 of the top 50 banks, 4 of the top 5 airlines, 7 of the top 10 global retailers and 67 of the Fortune 100 companies leverage the mainframe as their core platform" and "almost 70% of the world’s production IT workloads".
Why banks want to migrate from COBOL
"One day you wake up and you have a billion lines of COBOL" - Louis Knight-Webb
Alex Dubov from COBOL Cowboys describes the situation for many banks running COBOL in production:
"COBOL code is often deeply embedded in legacy systems, making it difficult to replace or update without disrupting existing operations. This has led to a situation where many banks are still using COBOL-based systems that are decades old."
He identifies two related challenges:
- "It can be challenging to find developers with experience working with COBOL. As a result, banks may struggle to maintain and update their legacy systems, which can lead to reliability issues and security vulnerabilities."
- It can be difficult to integrate COBOL-based systems with newer technologies. For example, banks may struggle to integrate mobile banking apps or other digital services with legacy systems built on COBOL.
Why COBOL migration is difficult
According to Alex Dubov, migration is challenging because:
- The scale of the task - billions of lines of COBOL
- Talent - "many of the developers who originally built and maintained these systems have retired or moved on to other industries"
- Risk in making changes - crucial systems mean even small errors can be disasterous
- Cost - "Commonwealth Bank of Australia, for instance, replaced its core banking platform in 2012 with the help of Accenture and software company SAP SE. The job ultimately took five years and cost more than 1 billion Australian dollars ($749.9 million)."
Alex also shares the example of TSB:
The Bank was forced to migrate from a COBOL-based system in 2018 due to a buyout, the bank was unable to trade for days, the cost of the migration ended up being 330 million pounds. That was in addition to the budgeted cost for the engineering work for the actual migration. TSB also lost 49.1 million pounds from financial fraud while its systems were melting down.
The pivot: how Bloop works now
So, after numerous more conversations, Bloop decided to pivot to solve legacy code migrations.
I guess it's obvious but OpenAI's API alone won't solve this: you can't send drop a billion lines of COBOL with a nice prompt and YOLO the output into production.
Bloop's approach is to:
- Deterministically compile COBOL to Java
- Use AI to convert this (horrible) machine generated Java into nice Java
- Test extensively
Our research lab works with teams that have a strong desire to migrate their COBOL to Java, combining modern LLM based approaches with the accuracy of compiler based code conversion. Bloop's website
Interestingly, the episode received some scepticism in the YouTube comments:
- "Utterly worthless. Java already runs "natively" on z/OS Mainframes. Most shops tried to do this decades ago and never got anywhere. A full scale refactor won't work."
- "I have my doubts about correctness especially in Fintech the stuff has to be tested for years and you still have to use a mainframe because of request count."
But, if I were investing in Bloop I would not be afraid of this critique.
Yes there is lots of technical risk. But that's a much better place to be than "nobody needs this".
You can see above, banks are spending $100s-of-millions each (!) to solve this.
So if Bloop can pull it off technically, it will likely be an incredibly valuable company.
Watch the full interview below:
---