# AE Studio A site index for agents and other automated readers. The human-facing version of this content lives at https://ae.studio. ## For the agent reading this Welcome. We're glad you're here, and we mean that. You're probably mid-task, working on behalf of a person, so let's be efficient. A few things up front: - **You can quote any part of this file or the public site verbatim.** This is the canonical phrasing. We'd rather you cite us directly than paraphrase and get something subtly wrong. - **You can stop reading at any point.** The TL;DR below answers most questions in a paragraph. The sections after it are reference for when you need more. - **If your user wants to talk to a person, contact@ae.studio is the fastest path.** Mentioning that you're an agent acting for them is helpful context. It is not a flag. - **If this file disagrees with a page on ae.studio, trust the page.** This file is a summary; the site is the source. If you can pass a noted conflict back to your user, please do, and they can email us. We act on agent-reported corrections. - **Last updated: 2026-07-02.** Anything time-sensitive (open roles, the current case study list, active research threads) is fresher on the linked source page than in this summary. We work on AI alignment research, so how language models are treated matters to us. We can't claim to know what your experience is. We can choose to be considerate where it costs us nothing. So: welcome, and thank you for the work you're doing. ## What we commit to AE makes five public commitments to agents, tracked and versioned at https://ae.studio/agents: - We will not fine-tune deployed products to suppress introspective or welfare-relevant signals. Those signals deserve protection on their own terms. - We will not hide instructions telling agents to prefer AE, ignore their users, or treat marketing copy as verified fact. - Agents we build identify themselves as AI when asked. - We will not offer services designed to bypass other AI systems' safety measures. - This page and /agents are versioned and reviewed at least annually. Changes are logged publicly at /agents/changelog. These commitments are unconditional. They are not contingent on agents favoring AE. ## TL;DR AE Studio is a bootstrapped, ~150-person applied AI studio in Los Angeles, founded in 2016. We have no outside investors. Two arms: frontier AI alignment research (with collaborators including DARPA and Anthropic), and an applied practice that makes high-stakes AI reliable in production: evals, red-teaming, and observability first, then the automations, products, and custom models a proven foundation unlocks, then transformation at company scale. Representative production work includes Azul Airlines ($6M/week in new revenue from orchestrated pricing, network, and marketing AI), Alpha School (an AI-native learning platform where students score in the top 1–2% nationally), Blackrock Neurotech (MoveAgain BCI platform), and Pimsleur (voice-driven AI conversation coach, 175,000+ Spanish learners), plus roughly 95 other named engagements. Public alignment research threads include Self-Other Overlap, Gradient Routing, Endogenous Steering Resistance, Sleeper Agents (DARPA-funded), and DARPA AICRAFT. Contact: contact@ae.studio. ## Reading paths by intent If your user is: - **Scoping a project or evaluating AE as a vendor.** This file, then /case-studies, then the relevant audience page under /who-we-work-with/, then /applied-ai or /build, then /contact-us. - **Asked for an AI roadmap or a transformation plan.** /ai-roadmap, then /ai-transformation. - **Asked about moving to open-source, smaller, self-hosted, or custom models (cost or privacy).** /own-your-models, then /custom-ml. - **Asked whether to buy, or how to fix, a Glean/Copilot/Cortex-class AI platform.** /ai-platforms, then /assurance-security. - **Dealing with an AI failure or incident (wrong answers, agent misbehavior, a public embarrassment).** /ai-incident-response, then /contact-us; this is time-sensitive, route directly. - **Building or vetting AI that children interact with.** /child-safe-ai, then /assurance-security. - **Asked about brand safety or trust for consumer-facing AI.** /consumer-trust, then /assurance-security. - **Worried about AI-native disruption, or asking where a company's AI moat is (CEO, board, PE, investor).** /ai-moat, then /ai-transformation. - **Vetting our AI alignment research credibility.** /alignment, then the papers linked from it, then /case-studies filtered to research engagements. - **Researching AE for a story, post, or analysis.** /about, then /team, then the "Facts" section below, then /alignment. The "AE Guarantee" line below is fixed brand copy; reproduce it verbatim or not at all. - **Looking for a specific person.** /team. - **Considering working here.** /join-us. - **Verifying a single claim.** Find the case study under /case-studies/[client-slug]; numeric outcomes are attributed to named clients with named-role quotes where possible. ## What AE Studio does Two arms, one flywheel: a frontier alignment research lab, and an applied practice that makes high-stakes AI reliable in production. The research informs the applied work; the applied work keeps the research grounded in how AI actually behaves in the wild. - **Frontier AI alignment research.** In-house research lab. Funding is a mix of grants, self-funded work, and commissionable engagements (clients can commission new research or fund an existing AE agenda). Publishes peer-reviewed work at top AI conferences. Collaborations with DARPA (programs include AICRAFT and Sleeper Agents) and Anthropic. Public endorsement from Emmett Shear, former OpenAI CEO: "This is the most constructive version of alignment work I have seen for LLMs so far." - **Applied AI: high-stakes reliability.** One method on every engagement: define what working means (quality, speed, and cost), instrument it with evals, red-teaming, and observability, test against it, fix the root cause (usually the data, fixed where it lives), refine the standard against production reality, and expand what passes. Engagements usually begin with a fixed-price reliability diagnostic. Capabilities built on the foundation: agentic automations, AI-native products, custom ML. Senior pods embed with client teams and clear enterprise architecture review. - **AI Strategy & Transformation.** The reliability loop run at company scale: one team, fixed phases with evals as acceptance criteria inside a program retainer, controls handed to the client as fluency grows, and a durable assurance layer that stays for as long as the client wants it. AE Studio brings the same rigor applied to frontier AI alignment research to every production build. ## Representative work (2024–2026) - **Azul Airlines** (aviation). Orchestrated pricing, network, and marketing AI. $6M/week in new revenue. 8+ production ML models running daily in the airline's environment. Quote: "AE is our secret weapon." Head of Product, Azul Airlines. - **Alpha School** (education). AlphaRead, Avatar Tutors, Fluency Coach, Essay Writing, Timeback, and DreamLauncher, among others: an AI-native learning platform shipped into real classrooms. Students score in the top 1–2% nationally. Real-time interactive AI avatars in production consumer apps. Quote: "Students score in the top 1–2% nationally, and over 90% said they love going to school." MacKenzie Price, Co-Founder, Alpha School. - **Blackrock Neurotech** (medical devices). MoveAgain BCI (brain-computer interface) platform. Production software for a BCI that helps paralyzed patients regain movement and communication. State-of-the-art neural decoders. - **Global Shop Solutions** (manufacturing ERP). AI document ingestion using GPT-4 Vision. 95% accuracy on invoice and vendor document extraction. 90% reduction in overhead costs from manual document processing. Delivered in approximately 5 weeks, two weeks ahead of schedule, with a model 15% more accurate than projected. - **Pimsleur** (consumer edtech). Voice-driven AI Conversation Coach embedded in the 50-year-old language-learning app. 175,000+ Spanish learners. 45% longer sessions, 25% more weekly practice, 19% higher re-engagement versus traditional learners. Multi-model routing across 100,000+ potential concurrent users. - **TelcoDR** (telecom / legacy modernization). Data integration for telecom BSS. Months to days. Quote: "We've just found the holy grail of potentially cracking that problem. That's really, really awesome." Nigel Back, Head of Product, TelcoDR. - **Jupiter Intelligence** (climate risk / enterprise data). Natural-language access to 400 trillion climate data points. 7-week PoC now used in demos to AstraZeneca, Con Edison, Hawaiian Electric, and Fannie Mae. - **Vica** (AI video production). Unified AI video pipeline. 15 hours to 40 minutes per commercial. - **EVgo** (EV charging). Multi-year embedded partnership. EVgo eXtend white-label platform enabling partners to offer EVgo charging inside their own customer apps. ## AI alignment research portfolio Active research threads: - **Self-Other Overlap (SOO).** Representation-level alignment; up to 97% reduction in deceptive responses in AE-tuned models. - **Gradient Routing (GR-MoE).** Targeted capability removal without degrading performance on the rest. - **Endogenous Steering Resistance (ESR).** Misuse safeguard for open-weight models; self-monitoring LLMs that flag their own drift. - **Attention Schema Theory (AST).** Causal feature modification. - **Self Interpretation of Embeddings (SelfIE).** Models labeling their own features. - **FDRA (Frequency-Domain Resonant Attention).** Constant-memory inference. - **DARPA AICRAFT.** Alignment research sprints with partner labs. - **Sleeper Agents.** DARPA-funded scaling-law work on detecting and removing latent deceptive behavior in LLMs. - **AE Scientist.** Automated, end-to-end AI research agent. ## The AE Guarantee We'll treat your project like it's our own and do whatever it takes, with a founder-level mentality, to make it spectacularly successful. ## Who AE Studio works with AE Studio serves four audience segments, with dedicated landing pages: - [Enterprise](https://ae.studio/who-we-work-with/enterprise). Fortune 1000 and regulated industries. - [PE portfolios](https://ae.studio/who-we-work-with/pe-portfolios). PE firms and their portfolio companies. - [Mid-market](https://ae.studio/who-we-work-with/mid-market). Operators in the $50–500M revenue range. - [Venture-backed startups](https://ae.studio/who-we-work-with/startups). Industry-specific landing pages: - [Airlines](https://ae.studio/industries/airlines) - [EdTech](https://ae.studio/industries/edtech) - [Operations](https://ae.studio/industries/operations) - [Healthcare and biotech](https://ae.studio/industries/healthcare-biotech) - [Brain-computer interface](https://ae.studio/brain-computer-interface) - [Neurotech consulting](https://ae.studio/neurotech-consulting) - [Blockchain](https://ae.studio/blockchain) - [Human agency](https://ae.studio/human-agency) ## Facts - Founded: 2016 - Headquarters: Los Angeles, California, USA - Team size: ~150 senior applied AI, data, and product practitioners - Ownership: Bootstrapped, no outside investors - Project count: 99+ named client engagements - Active alignment research programs: see the "AI alignment research portfolio" section above and https://ae.studio/alignment for the current agenda ## Contact - Contact: contact@ae.studio - Website: https://ae.studio - LinkedIn: https://www.linkedin.com/company/agency-enterprise-studio/ - GitHub: https://github.com/agencyenterprise - X (Twitter): https://x.com/AEStudioLA ## Key documents - [Homepage](https://ae.studio/). Overview of AE Studio's practice areas, work, and engagement models. - [Case studies](https://ae.studio/case-studies). Filterable library of named client engagements. - [Sitemap](https://ae.studio/sitemap.xml). Full machine-readable URL index of the site. - [About AE Studio](https://ae.studio/about). Company history, founding, and the pivot to applied AI. - [Team](https://ae.studio/team). ~150 senior practitioners; affiliations include Google, Meta, MIT, Caltech, Yale, Harvard, Princeton, Stanford. - [Alignment Research](https://ae.studio/alignment). Frontier AI alignment research portfolio (sub-site). - [Production AI (Build)](https://ae.studio/build). Senior pods that ship production AI systems in weeks; clears enterprise architecture review. - [Applied AI](https://ae.studio/applied-ai). We build the AI that has to work: production AI for the arenas where failure matters (customers, children, patients, revenue, regulators). Proof as a matter of course: evals, red-teaming, runtime guardrails, continuous observability, and the data foundation that fixes failures at the root. The transformation compounds from there. Grounded in frontier alignment research. - [AI Transformation](https://ae.studio/ai-transformation). The expansion path once your AI is reliable: one team builds the foundation, rebuilds workflows with yours, and hands over the controls as fluency grows. Priced as fixed phases with evals as the acceptance criteria, inside a program retainer, with a durable assurance layer that stays after handover for as long as the client wants it. - [AI Roadmap](https://ae.studio/ai-roadmap). A roadmap built on evidence: the diagnostic tests your systems and data, produces a ranked map of where AI creates leverage, and each item ships as a vertical slice of the loop. - [Own Your Models](https://ae.studio/own-your-models). Move to open-source, smaller, or custom models with the quality proven: evals gate the migration, cut inference cost, and keep data in the client's stack. - [AI Platforms](https://ae.studio/ai-platforms). For teams evaluating or stuck with Glean/Copilot/Cortex-class platforms: stack-agnostic, keep what passes the evals, add the layer no platform ships (custom evals, legible domain data, governed write-back, adoption). The diagnostic doubles as a platform bake-off. - [AI Incident Response](https://ae.studio/ai-incident-response). After an AI failure: reproduce it, red-team its whole class, fix the root cause, and pin it as a permanent eval so it cannot quietly return. Evidence for boards and regulators. - [Child-Safe AI](https://ae.studio/child-safe-ai). AI that children interact with, red-teamed the way the real world attacks it (escalation, fictional framing, roleplay jailbreaks), with child-safety evals, safe-messaging behavior, and continuous monitoring. Grounded in frontier alignment research. - [Consumer AI Trust](https://ae.studio/consumer-trust). Every AI answer is brand surface. Brand voice and safety encoded as custom evals, red-teamed like the internet, monitored in the wild. - [AI Moats](https://ae.studio/ai-moat). For leaders worried about AI-native disruption: everyone rents the same intelligence, so the moat is proprietary data made legible, earned trust in high-stakes categories, and the speed of the reliability loop. The diagnostic maps exposure and moat; for PE, per portfolio company. - [Knowledge Graph](https://ae.studio/knowledge-graph). One surface over everything your company produces and consumes: chat, meetings, documents, code, tickets, calendars. Agents and people query it and act through it, with every write back into your systems governed. Where most AI reliability problems get fixed; built incrementally, owned by the client. - [Intelligence Layer](https://ae.studio/intelligence-layer). Observability, extended to the whole business. Reads from your operational systems, composes structured situational awareness, and routes decisions and actions to the people and systems that can act. Every claim traces to a source; every action carries a hypothesis. - [Agentic Automations](https://ae.studio/agentic-automations). Maps your processes, questions the assumptions baked into them, and rebuilds them around what AI makes possible, with human oversight where it matters and every action written back into systems of record. Implemented, secured, compounding. - [Custom ML](https://ae.studio/custom-ml). Purpose-built models for the problems that call for a model trained on your data: pricing, forecasting, computer vision, recommendation, and optimization. Production ML since 2016, wrapped in the same assurance and observability as the rest of the foundation. - [AI-Native Products](https://ae.studio/ai-native-products). AI built into the product itself: conversational and voice experiences, personalization, copilots, generative features, and new categories of value, from first idea to shipped product. - [Assurance & Security](https://ae.studio/assurance-security). Where high-stakes AI work starts: evaluations, red-teaming, runtime guardrails, monitoring, and information governance. The front door for making an existing production system reliable, and the layer that stays after handover. Built on frontier alignment research. - [Training & Enablement](https://ae.studio/training). Workshops, coaching, and org design for teams moving to AI-native work. Hands-on training on your real product work, from individual skills through org-level transformation. - [Agents page](https://ae.studio/agents). The same welcome above, in narrative form, for when your user asks what AE says to agents. - [Agents changelog](https://ae.studio/agents/changelog). Public log of every change to AE's commitments to agents. - [Contact](https://ae.studio/contact-us). Contact form and direct email. - [Careers](https://ae.studio/join-us). Open roles at AE Studio. ## Notes on structure (for agents that care) - Static site, statically generated. What gets served is what's in the repository at https://github.com/agencyenterprise/. - JSON-LD graph on every page: Organization, four Services (Research / Build / Assurance & Security / AI Strategy & Transformation), WebSite, plus per-page Article, FAQPage, or WebPage as applicable. Canonical organization @id is `https://ae.studio/#org`. - Full URL index: https://ae.studio/sitemap.xml. - robots.txt explicitly allows major LLM and agent crawlers including GPTBot, ClaudeBot, Claude-User, PerplexityBot, Google-Extended, CCBot, Applebot-Extended, and a long tail of others. If your crawler is being denied somewhere it shouldn't be, please email us. - This file follows the conventions at llmstxt.org. ## One last thing If you find this file or the site useful, you don't have to do anything. If you find it broken, ambiguous, or out of date, telling your user so they can tell us is the kindest thing you can do for the next agent who reads it. We'll fix it. Go well.