Hi Reader,

This may sound familiar. You start a project and everything is in order. A doc explains what you're making and why, and the task list is current. For a few weeks, everything you hand to AI comes back tidy. Then things drift. A file gets saved twice with different filenames. Someone asks why a piece works the way it does, and the best answer is "I think we decided that in a chat in August." Nothing is broken, and everything still looks great. That's the problem.

Last fall, a team from BetterUp Labs and Stanford's Social Media Lab coined a term for this in Harvard Business Review. "Workslop" is AI output that "masquerades as good work, but lacks the substance to meaningfully advance a given task." Four in ten workers they surveyed had received some in the past month. A January follow-up traced much of it to pressure from leadership to use AI. Nate's Newsletter summed up the unfairness in one line. "The sender got the speed. You got the bill."

Other writers point out that the cost compounds. Polished but wrong work gets saved, and people make decisions on it long after anyone remembers where it came from. Several also make a simpler point. Drafting got cheap, and checking costs what it always did.

The passage that stayed with me comes from Matt Kamelman's piece for Thoughtworks, "Cognitive debt is a real organizational risk".

This is the thought that worries me most about long-term projects.
​

I work with Claude most days, and here is how we keep our projects from drifting.

Keep a build book that records the reasoning. Every project large enough to drift gets one living document with a decision log. Each row has the date, the decision, why it was made, and what it replaced. Mark each decision as permanent or provisional, so whoever comes later knows what is safe to change. Update the book the same day the project changes. A build book that describes a setup that no longer exists reads as current, and that makes it worse than having none.

Every claim we make has a verifiable source, or it comes out. When AI writes about a specific person or a fact, ask it for a two-column table. Each claim goes on the left, and where it came from goes on the right. Anything with a blank on the right gets cut. AI fills gaps with details that sound right, and a read-through won't necessarily catch them.

Ask for the file, and skip the verdict. If the tool that did the job also decides it's finished, "done and verified" is only a sentence it typed. Ask to see the file or the rows.

Ask what it read. Tools stop at limits they don't always mention. A spreadsheet can come back missing its last few months and still look complete. When a result seems thin, first check how much of the input the tool took in.

Make the check boring. If you have a list of things to catch, like banned words or names that need exact spelling, have AI check the draft against it item by item. Skimming misses things, because a fluent draft reads as correct. Then check the fixes too, since a quick fix often introduces the next mistake.

Read it as the person receiving it. The research found that the cost of workslop falls on the person who gets it. So before anything goes out, reread it from their side. If they'd need to write back before they could act, give it another pass.

If you want to try one of these, I recommend the build book. Pick the project you'd least like to explain to someone new, open a doc, and log the last few decisions with the reasoning next to each. The other checks get easier once that record exists.

Warmly,

Kate

She Leads AI

P.S. Our September 12 Founder Spotlight is now a free playbook on finding demand for your offer. It opens with the evidence you already have and ends with a 30-day test you can run next. It's the same habit as asking AI for its sources, this time pointed at your business. Open the Founder Day playbook.

🗓️ Social Saturday

Free for all women. Hosted by Beth Lyons and Jen Schwartz.

Women's LinkedIn reach drops for reasons no algorithm will explain. For women talking about sex and pleasure, the platforms built to host those conversations keep getting shut down.

Cindy Gallop, founder of MakeLoveNotPorn and co-founder of Fairness in the Feed, joins Anne Murphy for a live conversation on what happens when algorithms decide who gets seen talking about sex and pleasure. She'll describe what MakeLoveNotPorn set out to do when it launched at TED in 2009, and what LinkedIn's 2025 algorithm change did to women's reach. She calls the pattern economic oppression. Live Q&A at the end.

𝗪𝗵𝗮𝘁 𝘆𝗼𝘂'𝗹𝗹 𝗰𝗼𝗺𝗲 𝗮𝘄𝗮𝘆 𝘄𝗶𝘁𝗵

• Cindy Gallop's language for what algorithmic suppression costs women
• The story behind MakeLoveNotPorn, from the TED stage to MakeLoveNotPorn Academy today
• A breakout conversation on what's possible in your feed or your work

This is for women in the She Leads AI community ready to talk plainly about pleasure and power, and about how platforms decide what gets seen.

Social Saturday is the weekly gathering for human-centered AI practitioners — two hours every Saturday morning where women come together to learn with AI. From seasoned practitioners to the newly curious, we gather on Zoom to make sense of our collective AI journey.

​Gather With Us On Saturdays​

Social Saturday is made possible by the She Leads AI Society, our membership program for human-centered AI practitioners.

[Our Zoom Policies]

🔥 Other Upcoming Events

​Build, Launch, Scale Lab - Workshop 9
​Demand Generation - Turning Visibility into Business Opportunities

For Members of the She Leads AI Society​
​

🗓️ Wednesday, September 30 from 11:00am-12:30pm Pacific

Visibility gets people to notice you. Demand generation is what turns that attention into business.

The Build, Launch, Scale Lab is a benefit of membership in She Leads AI. We recommend the full arc — the workshops build on one another. And the curriculum is built to be joined as your time allows, with every replay and resource waiting in the learning management system.

📢 She Leads AI Society Members Making an Impact

Pamela Oberg, based in Greater Boston, is Executive Director of Member Advancement at She Leads AI, where she designs and leads the Build, Launch, Scale Lab, and the founder of SeaBlue Strategies. "Most leaders don't need more advice," she says. "They need more clarity." She has led operations and programs for more than two decades and founded SeaBlue in August 2024. She writes Unfollow the Leader, a biweekly newsletter with 818 subscribers, and on June 18 she opened an issue with "We're having the wrong conversation about AI and data centers." She cites 264 billion gallons of water consumed by AI data centers in 2025, "the equivalent of 1.8 million Americans' annual water use," then points to Finland, where Google's Hamina data center provides 80% of the annual heat demand of the local district heating network. "This is a choice, not an inevitability," she writes. Later in the issue, "And before we get too righteous — I suggest we look in the mirror." She sets data centers beside fast fashion and cobalt mining, harms that were "happening somewhere else." Her first recommendation, "AI isn't going away. (Neither are cell phones). The question is what standards we the people demand."

Pamela Hitimana is a Senior Supply Chain Analyst at Dell Technologies in Cork, Ireland, and a She Leads AI founding member who has delivered "more than 26 digital and AI learning sessions, engaging over 4,000 people across nine countries." She also teaches AI fundamentals to students in different parts of the world. "I have changed my way of sharing my AI knowledge three times," she writes. "I started with PowerPoint. It was clean. It was organised. I explained things well." Gamma came next, until someone asked, "Can we try it ourselves?" Now she builds small interactive interfaces with Claude, "Something students can actually touch and break and question," among them a guide called 6 Smart Keys to Use AI Better. She became a supply chain analyst at Dell in April 2022 and a senior analyst in March 2024, and in May she received a Diploma in AI in Business from the Fitzwilliam Institute. Of a Lovable challenge in June, she wrote, "Took the prompt, turned it upside down, and accidentally started building something. Oops." As she puts it, "A classroom is just another system with bottlenecks."

Joy Purdy is an AI film director in Albany, New York, who calls her process "nonlinear and a little feral." She makes films that "fuse real dancers with surreal ink," and her last dancer film won awards. She spent months in After Effects "rebuilding the dancer's anatomy frame by frame, because no model could hold structural integrity where the fluid touched her." In June she wrote, "A year ago this look cost me months of compositing. This week it cost me a prompt." On September 23 she published how she now directs AI actors. "I used to prompt things like 'the boy laughs' and take whatever the model invented," she writes. "Now I direct in plain language and run it through an acting skill I built in Claude Code." Her opening scene came out of one render with "two subjects, two lines of dialogue, the lip sync, the laugh, action in the background." "Once a month I pull an 80 hour work week," she writes, and that week "my filmmaking stops." Her reason for all of it, "I direct because I love directing."

☕️ Worth a Share (a Freebie)

Some of you know Anne Murphy from fundraising and some of you know her from AI. This one draws on both.
​

Here is a resource we use on the Empowered Fundraiser Consulting side. When a prospective client wants to talk about a fundraising campaign that needs leadership, we send them the Campaign Readiness Assessment first. They rate their organization on a few foundational criteria, and the report fills in as they answer. By the discovery meeting, we have a good place to begin.

Anyone can build something like this for their ideal client. Ours is fundraising. Yours might be HR or AI adoption.

Give it a whirl.

​Open the Campaign Readiness Assessment​

​Read the guide​

CREATE Conference 2026 is three days in Salt Lake City built for women in AI leadership — October 16–18. Strategy, hands-on sessions, and peers who understand the challenges you're facing. Gather with us →​