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

Most of what a company knows already exists somewhere: in files, recordings, research reports and code. The smaller items this week turned on where that knowledge now runs, on whose servers, and whether the tools built to remember it help.

On Wednesday, at its FUSION conference in Las Vegas, the automation software company UiPath launched Cartographer, a tool that builds what it calls a Map of Work: a governed record of how processes actually run, including the exceptions, workarounds and judgment calls that never reach an official workflow document. It gathers that record by reading a company's existing documents, interviewing subject-matter experts and watching screen recordings. A Decision Ledger stores each exception decision and why it was made, and proposes it back into the map for a human owner to approve. (UiPath, Sep 23)

What's new: Until now, companies captured expertise by writing down the steps of a process, often in a documentation push before a retirement, and the reasons behind the steps left with the people who knew them. This week UiPath shipped a way to record the reason at the moment a person makes a call, Rockwell Automation published what a troubleshooting assistant built from veterans' knowledge did for new technicians, and a Korean refinery project began tying veteran judgment to its equipment records. For a company, the question moves from whether its processes are documented to whether it keeps the why behind its exceptions, and who checks it before anyone relies on it.

By the numbers: Most large industrial firms say they have not closed that gap. A company called Octave Intelligence, which sells industrial software, surveyed 756 technology decision makers at industrial firms with more than $1 billion in revenue in April and May. About seven in 10 said they are not retaining the knowledge that will matter most for future operations, and 47% said core processes are recorded while the know-how behind them is not. (Octave, Sep 23)

Between the lines: Rockwell Automation's plant in Singapore shows what a plant can do once veterans' knowledge is written down and searchable. Since October 2025, its technicians have used a maintenance assistant that searches a database built by veteran engineers with 20 to 30 years on the shop floor, organized by symptom, cause and response, so a technician sees the likely causes of a fault first. Li Wang, the plant director, said the aim is 'The loss of knowledge, what we call the tribal knowledge,' and that machine downtime fell 33 percent. Bob Buttermore, Rockwell's chief supply chain officer, said new technicians now learn to troubleshoot hundreds of machines in three months instead of nine, and the assistant will roll out to plants in Twinsburg, Ohio, Poland and Mexico. The account ran on Microsoft's news site; Microsoft supplies the underlying AI service. (Microsoft Source, Sep 24)

In South Korea, a project under the Ministry of Trade, Industry and Energy is trying the same thing at refinery scale. Enhans, an AI company, and the refiner GS Caltex are digitizing veteran engineers' judgment and linking it to 832,996 equipment records and about 2,900 unstructured documents, served to field staff through a guide chatbot. Lee Seung-hyun, Enhans's chief executive, said the aim is to 'go beyond simply documenting the experience of skilled professionals and transform it into knowledge that companies can continuously utilize.' (VentureSquare, Sep 23) In Japan, TV Tokyo on Thursday featured SHUGI, a system from a company called USEYA that records master craftspeople's movements and line of sight through smart glasses so trainees can follow along. (USEYA, Sep 17)

Yes, but: Capture depends on people who are willing to share. Amy Loomis, a research director at IDC, said the step 'will only be as strong as the tribal knowledge of teams and the ability to capture it accurately.' Andrea Simpson, who runs automation at USI Insurance Services, said 'So much knowledge is in a person's head,' and that some of it, such as a phone call or a conversation at a desk, her firm wants to keep human. Daniel Dines, UiPath's chief executive, conceded on stage that a first Map of Work 'will be incomplete for sure.' (Computerworld, Sep 24)

Between the lines: The practitioners quoted this week agree on what to capture. Gleb Tsipursky, a behavioral scientist, told SHRM that AI helps most against brain drain 'when it captures why experienced employees made decisions, not just what they did.' Surojit Chatterjee, chief executive of the AI company Ema, said a documentation sprint before a retirement wave 'will fail the same way every previous one did.' (SHRM, Sep 21)

The opening: Microsoft says the record itself is the asset. Kathleen Hogan, its chief strategy and transformation officer, wrote that the company's AI learning loop 'was also generating unique institutional knowledge, and that retaining control of it would be critical to our long-term advantage.' (Official Microsoft Blog, Sep 17) A research paper posted on Tuesday proposes the safeguard: each captured fragment of expert practice is checked by a group of people before any AI system may use it. Its authors say whether that improves operations 'remains an open empirical question.' (alphaXiv, Sep 22)

Hogan's point applies beyond Microsoft: a company's record of how its people handle exceptions belongs to the company, whichever AI model reads it later. Before anyone relies on that record, someone has to be named to check each entry, and UiPath, the paper's authors and Rockwell's engineers all put a person in that seat.

In August, AI agents made 257 million requests to documentation hosted by Mintlify, a documentation software company, against 131 million page loads by people, according to its State of Knowledge report published last week. Agents were also directed to update those documents about 367,000 times between February and August. (Mintlify, Sep 16)

What's new: Until now, a company wrote its documentation for people, and a stale page misled whoever happened to read it. With agents now sending more traffic to Mintlify-hosted docs than people do, a wrong page gets repeated to everyone downstream, and only 7% of the 329 respondents Mintlify surveyed said all of their company's knowledge surfaces are readable by agents. The question moves from whether a company has written its knowledge down to whether that writing is accurate, current and cleared for the software reading it.

The big picture: Researchers from DX, an Atlassian subsidiary, with GitHub, Google, Capital One and the University of Victoria, published a framework in ACM Queue called CAFE(S) that gives teams a shared vocabulary for five dimensions of the context handed to AI tools: clarity, actionability, fidelity, efficiency and, set apart, security. The authors describe it as a way to discuss and evaluate context, not a measurement system. Brian Houck, a distinguished scientist at DX and co-author, said AI 'dramatically increases the cost of poor knowledge management. When agents receive poor context, they are forced to guess.' (ACM Queue release, Sep 24)

Between the lines: LinkedIn already keeps its procedures in a form its coding agents can read. Ajay Prakash, a LinkedIn software engineer, described a shared context layer that serves about 8,000 daily users and more than 600 playbooks of how the company does things; LinkedIn reports a 20% productivity gain. (InfoQ, Sep 19)

The opening: Research a company already paid for is starting to reach everyday AI tools. Stravito, a company that stores consumer research for its clients, launched a connector that lets ChatGPT, Claude and Copilot answer from that research with sources cited and each user's existing permissions enforced. Its chief executive, Thor Olof Philogene, said 'The most valuable consumer research is worthless if it never actually guides real business decisions.' (PR Newswire, Sep 24) Panopto, a video platform company, did the same for recorded video, citing research in which 68% of learning and training leaders reported that most of their recorded video training is not searchable. (PR Newswire, Sep 23)

Yes, but: A map of company knowledge built once goes stale. Andi Gutmans, who runs Google Cloud's data business, said 'Anyone who says, "I have a Knowledge Graph today" is trivializing the problem.' (diginomica, Sep 23)

Esther Kang, a senior lecturer at the University of Bristol Business School, ran five experiments with more than 1,000 adults aged 18 to 77, in which some participants formed social connections in simulated workplaces and online communities before learning new information. The more connected participants got better at remembering where knowledge lived and who held it, and less likely to retain the information itself, she wrote in the Harvard Business Review on Wednesday. She calls it 'the connectivity tradeoff.' (HBR, Sep 23)

What's new: Until now, many companies treated easier access as the same thing as learning: connect people, index the files, and knowledge would spread. Kang's experiments tested connections between people rather than AI tools, but she draws the lesson for AI-enabled knowledge systems: 'Access to information does not automatically translate into learning.' This week Atlassian and IBM added evidence from teams using AI. The design question moves from how fast an employee can find an answer to what the employee still knows without looking it up.

Why it matters: Atlassian's Teamwork Lab ran a brainstorming experiment with 797 knowledge workers. Starting with AI raised the number of ideas but lowered ownership and enjoyment, and the most original ideas came from people who reviewed a colleague's ideas before working alone. Molly Sands, who leads the lab, said 'starting with AI comes with a hidden cost: lower ownership and less joy.' (Atlassian, Sep 21)

By the numbers: The IBM Institute for Business Value surveyed 1,500 chief human resources officers and 8,800 employees between April and June; employees worried most about losing critical thinking. Amit Das, chief human resources officer of the Indian media group Bennett Coleman, put it this way: 'Fluency without judgment simply helps an organization make mistakes faster.' (IBM, Sep 21)

The practice: Beth Smith of The Cynefin Co, a consultancy, argues that complex, consequential work should stay human-led because 'capacity built through delegation isn't the same as capacity built through practice.' (The Cynefin Co, Sep 18) Leo Lo, dean of libraries at the University of Virginia, offers a working check: sort each AI answer as factual, interpretive, constructive or strategic, then check each kind differently, facts against sources and advice against your own circumstances. (Fortune, Sep 22)

In these studies, people kept learning where someone arranged the work for it: Atlassian's most original ideas came from people who read a colleague's work before opening an AI tool, and Smith would keep consequential work with people so they keep practicing it.

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 110 candidate items across seven source classes: journalism, vendor and company announcements, practitioner publications, research papers, government and institutional records, event programs and registries. 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.