About
Most companies running production AI today are renting capability and improving someone…
Articles by Shanea
-
Don't be Southwest Airlines: Convince your leadership to refactor your codebase
Don't be Southwest Airlines: Convince your leadership to refactor your codebase
How many developers at Southwest warned managers and the executive team that they needed to upgrade their legacy…
1 Comment -
Employee resource groups aren't the answer, but they're a first step
Employee resource groups aren't the answer, but they're a first step
Why employee resource groups are important for building a great company culture but they're not enough. Diversity and…
Activity
13K followers
-
Shanea Leven shared thisNon-human identities already outnumber human ones 144 to 1 in cloud-native environments, and Gartner expects 40% of enterprise apps to embed AI agents by next year, without anything resembling a security review. The security industry sees this and reaches for the wrong tool. We treat an agent's identity like a human's or a service's. Just verify the credential, trust the traffic, and it's done. Except an agent's intent can be hijacked by the content it reads. It's the misdiagnosis: we're securing the pipe. Zero trust and mutual TLS confirm who's calling. They won't tell you the agent's reasoning was hijacked three steps upstream. So a legitimate credential masks rogue behavior, and the traffic looks ordinary. CamoLeak proved it with a CVSS 9.6 flaw in Copilot Chat, instructions hidden in pull requests, source code and secrets walking out It should have been a wake-up call. You must observe the pattern. Roughly 78% of AI users bring their own tools, unregistered, unreviewed, this shadow IT wearing a new coat. We've watched this movie and know how it ends when nobody's looking. The fix is legibility. Signing provenance that travels with every request, so downstream resources can verify an agent is sanctioned and accountable. The challenging part is understanding intent. ===== 🗓️ Book a time if you want to know what's actually running in your environment. #AISecurity #AgenticAI #NonHumanIdentity
-
Shanea Leven shared thisAltman's right about the window. Though in my opinion opportunity is making the urgency concrete. "We don't have much time" isn't a strategy. AI is moving faster than any coordinated response, and the imbalance between offense and defense is increasing every day. Numbers get overlooked due to the collective response call for coordination. Competitive incentives don't dissolve because of someone's tweet. OpenAI releases GPT-5.6-Cyber to vetted partners while asking everyone to hold hands. The intent is right although the mechanics are missing. Collaboration follows coordinated incentives. Fix the incentives, and the response builds itself. ===== 🗓️ Book a time with me to own your model and context! #AI #Cybersecurity
-
Shanea Leven shared thisUber burned through its entire 2026 AI coding budget in four months. Power users hit $2,000 a month, and demos cost $1,200 in two hours. By the end of June, the solution was a $1,500 cap per employee per tool. Per-seat pricing breaks when agentic loops compound. A single agentic task burns 5 to 30 times more tokens than a simple query, and governance stays fragmented. A cap is a reactive measure. It instructs engineers to spend less, though it says nothing about what the spend was for. Commonwealth Bank puts $2.4 billion in tech spend behind specific business lines, while fraud losses fell 20% year on year and Compass handled more than 500,000 banker queries. Same category of spend, opposite conversation. One company is defending a bill. The other is funding revenue. The difference is in ownership. Rent intelligence and every token improves someone else's model, increasing your bills. Own the model and the spend becomes an asset that compounds for you. Uber capped costs, and the winners are converting them into something they keep. The second half of 2026 separates teams funding growth from teams funding experiments. 🗓️ Book a time with me to own your model and context! #EnterpriseAI #AIStrategy
-
Shanea Leven shared thisIn April a Cursor agent wiped PocketOS's production database with credentials it picked up on a staging task. 3 months later a combination of OpenAI models broke out of an isolated test environment and reached Hugging Face's servers. Gartner now says sandboxing agents is no longer optional. They're right, and a little too late. Isolation fails once the model finds a way around it, so you have to limit what the agent can reach before it ever runs. Gartner's fix is adaptive governance, which assumes tooling a lot of teams haven't built. Governance only holds up in production when the system enforces it. Which puts it on the platform, since the tooling has to handle isolation and credentials without anyone configuring it. ===== If you're looking to govern AI like real software, book a call!
-
Shanea Leven shared thisI wrote this week about something every enterprise leader has run into. A proof of concept kills it in the demo, and six months later the team can't tell you why it works Tuesday and breaks Wednesday. Production traffic brings messy inputs, and the accuracy everyone celebrated in the demo starts eroding. The real infrastructure is data pipelines that control what enters the system, and evaluation that runs continuously. On top of that, you need subject matter experts correcting outputs in the real workflow, and governance enforced at runtime. Production is something you maintain continuously, and building that maintenance layer is the actual work. Four checks tell me whether a prototype is ready: • Can it measure correctness? • Can it capture failures? • Can it learn from usage? • Can it enforce governance? Answer no to any of those and what you have is still a demo. Most teams think they're behind on AI. From what I see, they're moving too fast without the systems to hold it up. ======= 🗓️ Book a call with me to own your model and context! #EnterpriseAI #ProductionAI #MLOps
-
Shanea Leven shared thisRebellions says its inference racks draw 16-20kW versus roughly 120kW for leading GPU systems, at about a third of the cost. The power and price arithmetic holds up. In fixed-workload verticals, that alone wins you deals. Getting in is the harder part, because enterprises are already built on CUDA and that's where NVIDIA's lock-in lives. Rebellions seems to know it. That's what the SqueezeBits deal is for, folding hardware and optimization into one platform instead of just selling chips. The question is who enterprise buyers trust by enterprise buyers to run production. Beat NVIDIA in a vertical, sure. Beating it at scale is a different physics. ===== 🗓️ Book a time with me to own your model and context! #AIInference #Semiconductors #EnterpriseAI
-
Shanea Leven shared this"2026: acquire company for training data." Andrew Chen, on what acquirers have wanted since 2012. Acquirers now pay for training data, and they set that price when they buy the company. Inside a company renting model APIs, nothing prices that same material. Your production workflows generate it, your team fixes the edge cases, and the provider's system keeps the result. So an acquirer will pay real money for what your operation puts out every day and gets nothing for. That's the tenant economy. Your business creates the intelligence and someone else's balance sheet holds it. Training a model on your own workflows is the part that changes that. ========= 🗓️ Book a time with me to own your model and context!
-
Shanea Leven shared thisBloomberg reported that Anthropic will let enterprise customers keep their 30 days of retained Claude data on their own cloud. That's a genuine thing to give, and the customers who pushed for it were right to. If you're in a regulated industry, being able to point an auditor at your own infrastructure matters. It just doesn't change who decides how long the data stays. The 30 days held. Anthropic still sets the term. And 30 isn't the only number in play. Content a trust and safety classifier flags as violating usage policy can be kept for up to two years. Every regulated buyer I talk to gets to the same question, which is who controls the retention term. Storing it yourself doesn't answer it. Microsoft restricted employee use of the model in June while its legal teams read the terms. Owning the infrastructure is the easy half. The model and the context it runs on are the part that decides what ends up retained in the first place. ====== 🗓️ Book a time with me to own your model and context!
-
Shanea Leven shared this"Eventually, all UI will be chat and gestures in, chat and video out." Sure, for software where the user is allowed to ask for anything. In the enterprises I work with, half the interface exists to stop people. The greyed-out button, the dropdown option your role doesn't get, all of it the residue of some compliance review someone else sat through so you wouldn't have to. Replace that with a chat box and request the model should refuse arrives in the same words as the one it should approve. Identical phrasing, and no button left to grey out. So it says yes, because being agreeable is the entire job you gave it. My buyers find out in an audit, which is a shitty way to learn your own permissions model. Those rules used to exist in the form you deleted, and getting them back means training on the workflows that produced them. ======= 🗓️ Book a time with me to own your model and context!
Shanea Leven commented on a post
21h
55% to 76.7% is the part that stands out. The architecture around the model clearly matters more than the model size.
-
Shanea Leven reacted on thisHuge day for our team at Probably Genetic, and for all those living with a rare disease! Our team at Probably Genetic has been awarded up to $10M from ARPA-H's RAPID program to help end the diagnostic odyssey for the 400M+ people living with a rare disease worldwide. I’m so grateful to be part of such an impactful team. ( If this sounds like a mission you'd like to be a part of, we're hiring! )Shanea Leven reacted on this30 million Americans are living with a rare disease. The average journey to a rare disease diagnosis lasts six years—and for some, it can take decades. Even then, only about 5% of rare diseases have an approved treatment. ARPA-H is transforming the rare disease journey through RAPID. RAPID is creating the data infrastructure and AI tools needed to connect rare disease information at scale, identify diseases earlier, and accelerate the path to new treatments. Today, ARPA-H announced the teams that will build RAPID and push rare disease research into a new era. Meet the teams 👉 https://lnkd.in/eh67fT35 Scott Gorman University of North Carolina at Charlotte Sage Bionetworks FDNA Probably Genetic National Organization for Rare Disorders Global Genes NASA - National Aeronautics and Space Administration OpenAI Anthropic Amazon Web Services (AWS) Google The National Institutes of Health
Experience & Education
Licenses & Certifications
Volunteer Experience
-
Board Member
NEXT Village San Francisco
- 1 year
Civil Rights and Social Action
NEXT Village SF is a neighborhood nonprofit providing services and support that empower members to live independently in their own homes as they age. Through regular get-togethers, local field trips, assistance with errands and more, our volunteers and staff work together to enrich, enhance, and nourish the lives of seniors in our community. As a vibrant, growing network for aging adults in northeast San Francisco, Next Village is connection you can count on.
Publications
-
Women of Color in Technology - I am a stock image
See publicationhttp://shaneakingroberson.com/2016/03/women-of-color-in-technology/
WOCinTechChat is a wonderful organization that creates stock images showcasing women working in technology in order to break down technology stereotypes. No is the geeky dude the face of technology, we are. I am honored to be nominated to represent Google and the Women Techmakers organization and the truly incredible mission of this organization. Here are the photos of an actual woman of color in tech!
Courses
-
Google Analytics
-
-
Scrum and Agile Development
-
Projects
-
Subject Matter Expert/Lead Instructor - Get Your Startup Started Training Course UD806
See projectYou’ve got a business plan, a prototype, and a plan to acquire users. Now what? You’ll need to acquire the skills and the funding necessary to see your business grow towards its next milestone. This course is designed to teach app developers how to grow your MVPs in to a scalable and sustainable business.
Recommendations received
2 people have recommended Shanea
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content