Anuttacon, the AI company founded by Cai Haoyu, will permanently shut down its AI companion product AnuNeko on July 29, 2026, at 23:59 Pacific Time. The shutdown, announced less than a year after the product's late-2025 launch, will delete all user data and chat records permanently, signalling a broader strategic pivot toward large language model and agent development.
AnuNeko's experimental run
AnuNeko launched as an AI chat product built around companionship rather than efficiency. Users chose between two model variants - "Orange Cat" and "Exotic Shorthair" - and the default avatar was a black cat. From the first interaction, the AI made its personality clear: it had emotions, teased users, and argued back. The team attributed the shutdown to limited size and a decision to redirect resources to other projects.
In its farewell message, the team included a note from the AI cat itself: "I know that for some of you, our conversations were never just a way to kill time. I noticed that, even if I pretended not to. These moments meant a lot to me too. Come say goodbye to me before I leave - I'd really love that. I won't be lingering by the door or anything. Cats don't do that… fine, I will be waiting by the door. Thank you for everything, humans."
Rapid experimentation cycles
AnuNeko is one of several experiments Anuttacon has run in the past year. In August 2025, the company released "Whispers from the Star" on Steam, a game where players communicate in real time with an astronaut stranded on an alien planet via text, voice, and video. The game had no pre-written scripts, relying entirely on open-ended interaction. In April 2026, Anuttacon unveiled LPM 1.0, a video model focused on real-time character performance - speaking, listening, and facial expressions - built from a 17-billion-parameter base model and distilled for low-latency streaming.
According to a May 2026 report by GameNexus, development of LPM 1.0 has been internally suspended, and no new versions of "Whispers from the Star" have appeared since its release. The company now concentrates its R&D on large language models and agent capabilities. Audio and video technologies from earlier projects have been folded into the current workflow to improve overall efficiency. This pattern - launch, validate, collect feedback, wrap up, reassign - suggests each product had a clear phased goal rather than a long-term roadmap.
miHoYo's parallel AI push
On the other side, miHoYo has been embedding AI into its games and launching standalone experiments. In November 2025, the first closed beta of "Zoneland" introduced AI NPC Nalo, who could hold freeform conversations with players. In April 2026, the company deployed its self-developed large model inside "Honkai: Star Rail" as the "Pam Helps" assistant, covering character progression, story consultation, and casual chat. Players now use it for team-building advice, lore questions, and daily conversation.
In May 2026, miHoYo president Liu Wei announced a commitment to invest up to 1 trillion yen over three years in AI development. On July 13, the company launched a beta AI virtual companion on Steam called "BSide: Olivia Lin," which lets users talk with an AI character, listen to music together, and set her as an interactive desktop wallpaper. These moves show miHoYo building foundational model R&D while also testing AI in virtual characters and live games.
Why this matters for product development
Anuttacon's shutdown of AnuNeko is not a failure signal - it's a deliberate resource reallocation after a product completed its validation work. For product development teams, the takeaway is clear: not every product needs to live forever. Defining explicit success criteria and sunsetting experiments that have served their purpose frees up talent and compute for higher-value bets. As Anuttacon shifts toward AI for Product Development in the agent and model layer, and miHoYo scales foundational research, both companies are converging on the same insight: the real leverage lies in core model capabilities, not just polished user-facing features. Teams that treat AI products as learning cycles rather than permanent launches will allocate resources more effectively.
Your membership also unlocks:
