Humanity Labs gives wealth management firms an AI Workforce that makes growth profitable. In a people-based business, serving more clients means hiring more people, so cost scales with revenue. The AI Workforce breaks that trap.

It works inside the systems your team already runs and delivers done work across the front, middle, and back office, while your people stay focused on the decisions that matter.

Every job it runs writes into one shared memory, building the firm's Organizational General Intelligence (OGI) so it gets smarter as it grows.

The result is the Firm of the Future: a firm that serves more clients and grows without adding cost at the same rate.

Humanity Labs works with some of the world's leading wealth management firms, which together oversee more than $750 billion in client assets, and is backed by leading venture firms including Google Ventures and Vestigo Ventures.

Learn more at HumanityLabs.ai

Organizational intelligence is about keeping what an organization learns from its own work, but it's also about being able to prove it later. Two events in the news this week put the emphasis on that second part: a court that may treat a technically-preserved but unreadable record as no record at all, and the question of how far behind most employers still are on the oldest version of the problem, workers leaving before anyone captured what they knew. 

AI can now read a decade of code, tickets and change logs faster than any engineering team ever could. What it still can't do, Forrester analyst Biswajeet Mahapatra argued this week, is tell you why any of it is still there. 'A rule found in code may represent a valid business requirement, an obsolete policy, a workaround for a retired system, or a defect that has persisted unnoticed for years,' the article says. 'Only people can explain why it exists and whether it belongs in the future-state application.'

Why it matters: Mahapatra's argument is aimed at software modernization budgets, and it names a failure mode most of them fund only half of. AI can accelerate behavioral discovery, mapping what a system actually does. But it cannot substitute for intent discovery, interviewing the people who know which of those behaviors are load-bearing and which are debris. 'AI can accelerate each of those activities,' the article says. 'It cannot decide which parts of the past deserve a place in the future.' A company that skips the second kind of discovery doesn't find out until the rule it deleted turns out to have been a regulatory requirement, or the one it kept turns out to have been a workaround nobody still needed.

The stakes when nobody answers it in time: a federal court is already running the experiment. A judge in the Southern District of New York ordered OpenAI in May 2025 to preserve ChatGPT output logs that would otherwise be deleted after 30 days. News publishers led by The New York Times filed a sanctions motion in July 2026 alleging OpenAI kept deleting logs anyway and had, for more than two years, told the court it could not search data it had already built the capability to search. A 20-million-record sample OpenAI did produce came back so heavily redacted the court called it unusable, according to an August 2026 legal analysis. The company's own system did not preserve, in a form anyone could later explain, why it had done what it did.

The government hit the same wall from its own side: the National Archives told federal records officers on August 21 which uses of AI create a record the government is not allowed to simply delete. A records-management critique published nine days later, by Andrew Potter, a records-management writer who publishes as Meta Archivist, argued the memo answers the easy version of Mahapatra's question, custody, and skips the harder one, appraisal: what must survive so an AI-mediated decision can later be checked. NARA's own appraisal policy is unrevised since 2007.

Mahapatra's fix isn't a better scanner. It's a second, human step most modernization programs never fund. Behavioral discovery, what AI does well, tells you what a system does. Intent discovery, what still requires a person, tells you why, and it means finding the people who ran the system and asking them directly which rules are load-bearing before those people, or their access, or their memory of the reasoning, is gone. Programs that fund only the first kind get through the project faster and then spend years finding out what they deleted.

Where the same fix would help NARA: the National Archives says its memo draws its boundary narrowly on purpose. It states plainly that it 'does not establish policy related to AI governance, e-discovery, privacy, security, or ethical use.' That leaves the harder evidentiary question, what an agency must be able to reconstruct, to whoever writes the next policy, a job that now sits with Bradford P. Wilson, confirmed by the Senate on August 7 as the 12th Archivist of the United States.

What happens if nobody does either kind in time: the OpenAI case is what it looks like when neither behavioral nor intent discovery ran before a court asked. A preservation order that only says 'Keep it' does not, on its own, guarantee anyone can later explain what a system did and why. Proving that took a sanctions motion, and, as of this writing, still isn't resolved. That is the cost a private-sector operator is actually pricing when a modernization budget skips the interviews to save time.

Lance Warren's job at Burns & McDonnell is to sit an engineer down before they retire and get them to explain, on tape, decisions nobody wrote down the first time. Warren, the employee-owned engineering firm's corporate research librarian and archivist, described the practice in the July/August 2026 issue of Archival Outlook, the trade magazine of the Society of American Archivists.

Why it matters: the method is old and the deadline is not. APQC's Great Retirement research puts the current pace at 3.6 million baby boomers leaving the workforce a year, with 61 million gone by 2030, most of them without ever documenting the tacit expertise it took decades to build. 'The threat of the Great Retirement is more than a demographic shift,' said Lynda Braksiek, APQC's principal research lead for knowledge management. 'It's a wakeup call for organizations and C-suite leaders to act now.'

By the numbers: a 2025 survey of HR leaders found institutional knowledge loss was the single most common offboarding challenge, cited by 47 percent of respondents; 76.6 percent said they were concerned about it. Roughly 42 percent estimated that inconsistent offboarding, including knowledge loss, security gaps and rehiring costs, costs their company up to $500,000 a year. (- Enboarder, 2025)

What the interview method buys: an oral history is slower than a wiki page, and it captures something a form can't: the reasoning an engineer would only give if someone asked the right follow-up question. Burns & McDonnell keeps the recordings and transcripts in its corporate archive as institutional record, alongside the company chronologies and executive histories Warren also maintains. Nothing in the method requires AI. The AI-era pressure is what shortens the deadline, not what makes the interviews useful.

Elastic says it does not want to compete with OpenAI, Anthropic or Google. It wants to sit underneath all of them. In its first-quarter fiscal 2027 results, reported September 3, the company described itself as "the data substrate beneath frontier models" and said an expanded OpenAI collaboration would split the work: Elastic retrieves an organization's own documents, logs and communications, and OpenAI's models reason over what it finds.

Why it matters: that division of labor, one company owns the facts, another owns the reasoning, is becoming the default shape of enterprise AI, and it means the facts now have a market of their own. Three weeks before Elastic's results, a private equity firm called Oakley Capital paid up to £20 million for majority control of Graphwise, a knowledge-graph platform serving more than 200 enterprise customers with annual recurring revenue growing more than 30 percent. Oakley isn't buying a model. It's buying the layer that tells a model what a customer's data actually means.

The pattern started earlier this year: Redis launched Iris, a context and memory platform, in May, describing it as sitting between an agent and the business data it needs to act. Snowflake followed in June with Horizon Context and Cortex Sense, giving customers a shared definition of business logic that multiple AI tools can draw from instead of each reinventing it. Both address the same failure: an agent that reasons well but works from stale or ungoverned data isn't useful, whichever model powers it.

What this means for the people who run the system: an organization that buys a context layer is making a bet about which part of its AI stack it wants to own outright. The model can be swapped; a contract change can put a different one in the loop. But the record of what the business actually knows, who's allowed to see it, and where it came from is not something a vendor can swap out from under a company. That's the part organizations are now being asked to buy, govern and keep, on purpose, as its own line item.

This magazine, and all the magazines in the Humanity Labs AI Magazine series, were created using Hawkeye, an AI system built for people who work in fields too large and too fast-moving to keep on top of alone.

Every week Hawkeye reads thousands of items across mainstream press, trade publications, government records, vendor announcements and company newsletters, then clusters them to find the themes that are actually emerging rather than the ones already being discussed.

For this issue the consideration set ran to roughly 30 candidate items across five source classes: journalism, government and institutional records, vendor and company announcements, professional and trade publications, and litigation records. A first draft was written with Claude Opus, checked against primary sources, and edited by Marshall Kirkpatrick.

Where a claim rests on a company's account of its own product, this magazine says so on the page.