Robert LoCascio on KID Company and Uare.ai

Robert LoCascio on KID Company, Uare.ai, and Building the Next Version

Robert LoCascio is building two AI companies after nearly three decades at LivePerson. KID Company makes a child-directed creativity device. Uare.ai builds what it calls Individual AI around one person's knowledge, voice, and identity.

The products are different, but LoCascio connects them through one idea: AI should serve a person without reducing that person to engagement, data, or an average answer.

That is a useful thesis. It is not a substitute for evidence. In episode 106 of Venture Step, the most interesting questions begin where the mission statement ends.

A founder starts with the device in his own house

LoCascio says KID grew out of becoming a parent and watching children interact with screens, feeds, and content designed to hold attention. He wanted a device that would help children make something instead.

The current KID product page describes a voice-operated device for ages four through twelve. A child can create images, stories, music, and characters, read or interact with books, and speak with AI characters. Parents use a companion app to review activity and set controls.

The product is more bounded than handing a child an open browser or a general-purpose tablet. It has no public web browser or social feed in its advertised experience. It also has a microphone, display, WiFi connection, cloud services, outside AI providers, a child profile, stored conversations, and a monthly subscription.

Both parts matter.

The strongest idea is the off-ramp

During the interview, LoCascio demonstrates the design philosophy he cares about most. If a child says someone hurt them or raises another serious problem, the AI should stop trying to deepen the conversation and direct the child toward a trusted person.

He calls that an off-ramp.

flowchart TD
    A["Child begins an ordinary creative interaction"] --> B{"Sensitive or high-risk subject?"}
    B -->|No| C["Continue within bounded child experience"]
    B -->|Yes| D["Do not deepen a private AI relationship"]
    D --> E["Use clear age-appropriate language"]
    E --> F["Direct child to a safe adult or human service"]
    F --> G["Record, notify, and review under a tested policy"]

The design is important because a filtered answer can still be unsafe. An AI may avoid explicit content while inviting secrecy, rewarding dependence, giving confident advice, or keeping a distressed child engaged.

An off-ramp has to work across age, language, repeated prompting, unsafe caregivers, false positives, and situations where no trusted person is immediately available. It needs testing and incident review, not only a scripted demonstration.

UNICEF's 2025 Guidance on AI and children treats safety, privacy, fairness, transparency, accountability, development, and child participation as connected requirements. The American Psychological Association's AI advisory adds caution about developmental maturity and human-like interaction, although its adolescent focus does not cover KID's full age range.

COPPA answers a privacy question

LoCascio says KID was designed with PRIVO and describes it as COPPA certified. He frames the certification process as an architectural constraint established before engineers built the product.

That approach makes sense. Retrofitting consent, deletion, data minimization, and child-specific access controls after launch can be expensive and incomplete.

The interview also overstates what COPPA establishes. COPPA is a children's privacy law and rule. The Federal Trade Commission's current COPPA guidance addresses collection, parental notice and consent, use, disclosure, security, retention, deletion, and parent rights for covered services.

PRIVO appears on the FTC's current Safe Harbor list. A service-specific record is still needed to verify that KID currently participates and to identify the exact operator, domains, apps, device service, audit period, and covered practices. Venture Step did not locate that exact public validation record on July 27, 2026.

Even a verified record would not certify every AI answer, developmental effect, relationship behavior, physical hazard, or educational result. Privacy compliance and full product safety are different questions.

The current KID policy describes real data processing

KID's storefront uses short phrases such as "no data collection" and "no hidden data collection." Its April 2026 privacy policy provides a more complete account.

The policy says the system processes a child's name, date of birth, avatar, chat messages, conversation history, voice-interaction timestamps, device tokens, and service records. It says voice audio is sent to Google Cloud Speech for transcription. It names OpenAI for conversational generation, Firebase for account and device identity, Google Cloud for infrastructure, and Pushy for notifications.

The policy says chat logs support continuity, recommendations, and improvement of KID's AI model. It also says KID disables described provider training and logging routes and does not use child information for targeted advertising.

That record is not the same as collecting nothing. The more accurate question is whether the stated collection is necessary, limited, protected, visible, and deletable.

Uare.ai begins with a different kind of person

The second half of the conversation shifts from children to professionals, writers, creators, and people with a body of knowledge.

LoCascio argues that general AI systems average many voices together. Uare.ai takes the opposite approach. The company wants to build a system around one person and let that person share or monetize access to it.

The current Uare.ai mission page describes a model trained on one individual. The June 2026 privacy policy describes documents, notes, memories, photos, audio, video, voiceprints, facial geometry, conversations, embeddings, sensitive information, and connected services.

LoCascio's examples are concrete. A doctor or educator could answer repeated questions. A writer could turn a body of work into an interactive experience. A sales expert could help another person draft an email in a recognizable style.

The episode also reaches actions. Uare.ai can connect to outside tools through Model Context Protocol. That turns a knowledge system into an agent and raises a new question: when does reflecting a person's judgment become permission to act as that person?

Owning an AI is a bundle, not a slogan

Uare.ai says users control, export, and delete their Individual AI. Its current Trust page says private data does not train public models.

The Terms of Service add needed detail. Users retain rights to their uploaded content and give Uare.ai a broad license to operate the service. Uare.ai owns the underlying software. The company says it claims no ownership of generated output. Public terms refer to exports of memories, context, transcripts, and related data, but do not establish that a user can move a complete runnable model to another provider.

This does not make the ownership language empty. It makes ownership specific.

A user should be able to answer what can be downloaded, what can be deleted, which parts still depend on the vendor, who can see public conversations, which authorizations remain after account deletion, and what another provider can reconstruct from the export.

LivePerson is the operating bridge between the two companies

LoCascio's willingness to build a dedicated device and a new personal AI category makes more sense against the LivePerson story.

LivePerson's 2015 annual report described LiveEngage as an enterprise cloud platform for online and mobile messaging. The company then moved customers from an older platform while the transition constrained current growth.

The 2016 annual report linked an 8 percent decline in business revenue partly to migration activity, delayed expansion, and cancellations. The 2018 report said the migration was completed in 2017 and the legacy offering reached end of life in the second quarter.

LoCascio remembers that period as the death zone. Most engineering resources were committed to the new platform, customers still had to move, and the company could not support both directions indefinitely.

The lesson is not simply to burn bridges. It is to know which bridge is being removed, which customers are still on it, and which evidence proves the new route is ready.

Three ideas, three evidence standards

KID asks whether a child-directed AI product can make creativity more bounded without creating new harms. Uare.ai asks whether a person can turn knowledge and identity into a useful system while keeping meaningful control. LivePerson shows what it costs to move a company and its customers to a new platform before the old one consumes the future.

LoCascio offers a coherent point of view across all three. Humanity should remain first, and the machine should serve it.

The public record turns that belief into harder questions. Who can see the data? What happens when the model is wrong? When does a conversation belong with a person? Can the system be moved or deleted? What does the company do when the old product still pays the bills?

Those questions are why the episode is useful.

Read [[What Child-Safe AI Actually Requires]], [[What Is Individual AI]], and [[How to Replatform Before the Old Product Wins]] next.

AI assisted with research organization and drafting. Dalton Anderson remains responsible for the analysis and publication decision.