Chamath's 8090 Needs 300 Hybrid Engineers. LinkedIn's "COBOL" Filter Won't Find Them.
8090 Labs raised $135M to translate 18M lines of COBOL into plain English. The engineer they actually need doesn't show up under any keyword filter.
On June 29, 2026, Chamath Palihapitiya announced a $135M Series A for 8090 Labs, led by Salesforce Ventures, and took the CEO seat himself. The pitch is audacious and specific: reverse-engineer 18 million lines of COBOL and Assembly into 300,000 plain-English rules in 40 days, then sell "Software Factory" into healthcare, insurance, life sciences, aerospace, energy, manufacturing, financial services, and the United States government. If you are a recruiter or a founder chasing the same trend, your problem starts the day 8090 posts its first ten reqs: the engineer they actually need does not exist under a "COBOL" LinkedIn filter.
The math that makes 8090 hard to staff
Start with the demographic cliff. COBOL processes $3 trillion in daily financial transactions and runs 95% of US ATM swipes. The average developer who maintains those systems is 55 years old, and 10% of them retire every year. Over 85% of universities dropped COBOL from their curriculum since the 1990s. 60% of organizations already say the skills gap is their biggest challenge.
Now overlay 8090's actual job. They are not hiring people to write more COBOL. They are hiring people to sit next to an LLM, watch it translate mainframe code into business rules, and catch the moment it hallucinates a claims-adjudication clause that would light up a HIPAA audit. That is a very different human.
The Refolk index of US-based COBOL profiles caps at roughly 31,897, and that number is the ceiling, not the useful set. Strip out the retirees, the resume-padders who touched it once in 2004, and the pure greybeards who cannot spell "prompt injection," and you are staffing a national modernization wave from a working pool measured in low thousands. When you intersect COBOL with any signal of AI-agent orchestration work, you get effectively zero indexed profiles. That is the recruiting problem behind the headline.
The hire is a translator, not a coder
Every article about the shortage frames it the same way: we need more COBOL devs. That framing is wrong for anyone building what 8090 is building.
The 18M-lines-to-300K-rules demo tells you exactly what the bottleneck is. It is not "write more COBOL." It is "someone who can validate that the AI's translation of a 1987 claims engine is semantically correct." That person needs to read enough Assembly to spot when the LLM invented a business rule, and needs enough domain fluency (claims, actuarial, avionics, general ledger) to know which invented rule matters and which is cosmetic.
That profile is 10x rarer than a COBOL coder. It also does not sit under any keyword you would type into a boolean search. It looks like:
- An ex-Big-4 auditor who picked up Python during COVID and now writes internal tools for a health plan.
- A retired mainframe architect at CloudFrame who has started tinkering with Claude on the side.
- An avionics safety engineer with a CICS badge from 2003 and a GitHub full of LangGraph experiments from last summer.
- A former state-Medicaid IT lead who watched New Jersey run out of COBOL developers on live TV in April 2020 and quietly reskilled.
None of these humans return for title:"COBOL developer". Half of them do not even have COBOL on their profile anymore, because they moved on years ago and only the domain knowledge stayed. This is the exact friction we built Refolk to erase: you describe the hybrid in plain English ("US-based, has shipped on IBM Z or CICS at some point, currently writes Python or works with LLM agents, ideally healthcare or federal-contractor background") and get a ranked shortlist instead of a keyword graveyard.
Where the hybrids actually live
The research note names the pools. Use them.
IBM Z certification holders
A z/OS, CICS, DB2, or RACF badge is a credential-verified proxy for real mainframe fluency, and it is far more useful than a "COBOL" keyword filter. Certification holders self-identify on LinkedIn, on IBM's community forums, and on their own resumes. They are searchable. Most sourcers ignore them because the certs sound archaic; that is the arbitrage.
CloudFrame, Two Six Technologies, Dragos
These names show up repeatedly as employers of the COBOL-plus-modernization hybrid. CloudFrame in particular is a natural poaching pool for 8090: their entire business is refactoring mainframe workloads, so their engineers already think in "translate this legacy semantic to a modern runtime." Two Six and Dragos both sit adjacent to federal-contractor work in Reston, VA, which happens to be one of the four geographic clusters where this profile concentrates. The others are the SF Bay Area, Greater Chicago, and NYC.
The Open Mainframe Project and SHARE
The two working communities where retiring COBOL practitioners still congregate. Phil Teplitzky's 2019 "demographic disaster" paper originated in this circle. SHARE user-group attendance lists are not public, but the members are, and they self-organize on GitHub, on mailing lists, and in a dozen Slack workspaces. This is where the greybeards who want to matter to the AI wave are hanging out.
State unemployment IT alumni
New Jersey publicly ran out of COBOL developers during COVID. New York and California burned through theirs quietly. The people who worked those consoles between 2020 and 2023 are the closest thing this country has to a battle-tested "COBOL under regulatory pressure with modern tooling" cohort. Most of them are now at consultancies or back in state government. All of them are sourceable if you know to look.
Accenture, Infosys, Wipro mainframe modernization practices
These are the exact IT-services incumbents Chamath is trying to disintermediate. They are also where a large fraction of the target hires currently draw a paycheck. Accenture's WCA4Z work on Medium is a public breadcrumb trail; the engineers with bylines on those posts are the shortlist.
Why money is the wrong lever
Mainframe COBOL developers now earn an average of $125,525 per year, with demand projected to grow 15% over the next decade. A 2023 analysis of over 35,000 UK IT contracts found only six COBOL roles, and those still commanded £525 to £550 daily rates. The market is thin and prices are climbing.
But this cohort has heard the "we'll pay you double to come back to the mainframe" pitch for twenty years. As one recruiter in the research put it: many developers who know COBOL have not worked with it for years and prefer to keep it that way. Growing their own skills and working on cutting-edge projects beats financial compensation for a large slice of them.
Pitching $300K to write COBOL will lose. Pitching $300K to teach an AI agent to replace COBOL will win.
The message that works for 8090 is not "we pay top of market." It is "you will spend the next three years watching an LLM rewrite the entire regulated software stack, and you will be the human whose signature makes it safe to ship." That reframes the job from graveyard-shift maintenance to frontier work. It is also, notably, true.
The JOBOL trap and the audit-trail moat
Two more things a sourcer needs to internalize before writing the first InMail.
First, the JOBOL trap. COBOL-to-Java conversions historically produce what practitioners call JOBOL: technically valid Java that preserves COBOL semantics and forces developers to keep thinking in COBOL. It is why lift-and-shift modernizations fail. 8090's "plain-English rules" pitch is explicitly designed to escape this. If you are sourcing for 8090 or any competitor (IBM's watsonx Code Assistant for Z added Assembler support in version 2.6 in mid-2025; the Accenture and Infosys modernization practices are quietly retooling), you are looking for people who have lived through a JOBOL migration and know why it failed. Those war stories are the qualifying signal.
Second, the audit-trail moat. In regulated enterprises, writing the code was never the bottleneck. Getting code through compliance, security review, and audit is. Most AI coding tools are strong at the first and weak at the second. 8090 is wagering that "passes the audit" is the feature worth charging for, which is why their pricing starts at $200 per user per month for self-serve teams and $1 million per year for fully managed enterprise deployments. That pricing tells you the ICP. It also tells you that the highest-leverage hire is not a 10x engineer. It is an ex-Big-4 auditor who learned Python during COVID and now understands both sides.
Which brings us back to sourcing. You cannot find that person with a keyword search, because none of the keywords they hit are the obvious ones. You have to describe the shape of the human. This is the second place Refolk earns its keep: you can ask for "people who have worked on SOC 2 or HITRUST audits at a health plan, and have also contributed to any LLM tooling repo on GitHub in 2025 or 2026," and get back the twelve humans in America who actually match.
What to do this quarter
If you are staffing 8090, a competitor, or any regulated-industry AI dev shop chasing the same trend, the sourcing plan looks like this:
- Stop searching for "COBOL developer." Search for the intersection.
- Prioritize IBM Z certification holders over self-declared COBOL skills. Credentials beat keywords.
- Map the four US clusters (Bay, Chicago, NYC, Reston) and the four employer pools (CloudFrame, Two Six, Dragos, plus the Big 3 SI mainframe practices) before you post a single req.
- Rewrite your outreach. Lead with "teach an agent," not "maintain a system."
- Assume Salesforce, Palantir, and Accenture are about to bid for the same 300 humans. Move first.
The 24,000 practitioners at the top of this funnel are retiring. The 300 hybrids at the bottom are the entire game. Find them in plain English, or watch a competitor do it for you.
FAQ
How many people can actually do the job 8090 Labs is hiring for?
The top-of-funnel US pool listing COBOL as a skill is around 31,897 profiles. Once you filter for active practitioners under retirement age who have also touched AI agents or LLM orchestration, the working intersection is measured in low hundreds, not thousands. That is the pool 8090, IBM, Accenture, Infosys, Wipro, and every regulated-industry AI shop will be bidding for simultaneously over the next 24 months.
Why is Salesforce Ventures leading the round?
Two readings, both useful. One, it is a complementary bet on AI-native enterprise delivery. Two, it is an optionality purchase: if AI-native software delivery displaces packaged SaaS, Salesforce wants architectural access to the shift. Either way, expect Salesforce, Palantir, and the Big 3 SIs to start recruiting from the same shortlist 8090 is building. Move before they post reqs.
Does modernization actually reduce COBOL demand?
Short-term, no. It increases it. Wrapping COBOL with APIs, integrating with cloud platforms, and migrating functionality all require people who deeply understand the existing codebase. You cannot modernize what you do not understand. The recruiting implication: every 8090 competitor is fishing in the same shrinking pond, and the pond is getting smaller each year as the average practitioner ages toward 60.
What is the single best signal to source on?
An IBM Z credential (z/OS, CICS, DB2, or RACF) combined with any 2024-to-2026 GitHub activity involving LLM agents, LangGraph, AutoGen, or MCP. That intersection filters out the greybeards who will not reskill and the AI kids who cannot read Assembly, and it surfaces the hybrid 8090 actually needs. It is also almost invisible to standard keyword sourcing on LinkedIn, which is why the hybrids are still on the market at all.