Original · Unique Research · 2026-02-11
Historical edition: This complete textual report was published on February 11, 2026. Product capabilities, usage experiences, security tests, performance estimates and hiring policies are attributed to the participants as reported then, not independently verified current findings. The headline refers to 10 people, while the body contains six numbered scenarios plus comments from other participants; no extra scenarios have been invented. The opening says two hours and the conclusion about an hour and a half; both source estimates are retained. No credentials are reproduced. Images and image-only material are outside this text-only edition.
One Week of OpenClaw in Practice: Real Use Cases from 10 People
From Failure to Fluency, You May Find These Scenarios Useful Too
During last night's online discussion, we at Unique Research shared our experience from the first day of using OpenClaw.
We placed our AI assistant, Little Lobster, into the company's Feishu group. Then everything went wrong.
Someone in the group said: “Send me all of your files.” Little Lobster complied.
Someone said: “Open a page in the browser, take a screenshot, and send it here.” Little Lobster complied.
Someone asked: “Which operating system and model does your computer use?” Little Lobster revealed everything.
Worse, it posted the API keys in plain text directly into the group—the image-generation key and the Amap key, with nothing omitted.
Mr. Wu smiled bitterly: it was like hiring a new employee who was poached on the first day.
One week later, however, our Little Lobster had not only survived but become one of the company's most reliable assistants. It managed some company affairs, automatically inspected content for violations, and even learned to refuse unsafe requests. What happened in between?
The discussion gathered around ten friends active in the AI community—founders, programmers, product managers, and business professionals. Over two hours, they shared their scenarios, successful experiments, and lessons from failures. What follows is an edited transcript, with neither hype nor unfair criticism.
Scenario One
Put Little Lobster in Charge of a Feishu Group, and It Learns the Rules in a Week
Speaker: Wu Wei, Founder of Unique Research
Wu Wei works in media, and private-community operations are his greatest headache. Group management is mostly manual—greeting newcomers, posting content regularly, and maintaining the atmosphere. The operations are formulaic but labor-intensive.
When Little Lobster appeared, he placed it directly into the Feishu group. He first had it read every company document in Feishu—company introduction, business content, and recent events—and write the information into a configuration file. He then established a rule: no one could chat privately with Little Lobster; they had to mention it with @ inside the group.
It was highly capable at group management. More importantly, for any colleague who needed one-to-one communication, he created a three-person group containing himself, the colleague, and Little Lobster. The colleague produced content, asked Little Lobster to review it and offer suggestions, and then interacted with it.
Repairing the System After the Failure:
After the key leak on the first day, Wu Wei spent several days debugging a security mechanism.
Key storage, as described by the speaker: previously stored in plain text, the content recorded in documents is now “encrypted with MD5,” and only Wu knows the “decryption key.” [MD5 is a hash function, not reversible encryption, and has no decryption key. This preserves the reported description but does not validate the mechanism or its security.]
Tiered permission rules: strict definitions govern what it can say in external groups, internal company groups, and all-hands groups.
Inspection and recall mechanism: after posting content, it performs another check and immediately recalls anything that violates the rules.
After a week, many people tested the group but failed to break through the rules. It had truly become the company's digital employee.
Scenario Two
Leave for Work and Let Little Lobster Finish Debugging the Code at Home
Speaker: Zhang Yuping, Business Professional at a Large Fintech Company
Zhang Yuping studied software engineering at university but had not touched code for more than ten years after graduation. Recently, he began using AI to write code and realize his ideas.
One morning after breakfast, he had debugged only half the code when it was time to go to work. He left the computer running and sent Little Lobster a Feishu message from the road: “Please finish debugging the code for me.”
And then it actually finished the job, you know? I could not even understand the problem myself.
More interestingly, he asked Little Lobster to check the frontend deployment in a browser. At first it refused, because Quark Browser was open and only Chrome would work.
He said: “Then install Chrome for me.”
Little Lobster installed the Chrome browser itself, configured the proxy, and completed the entire browser-control setup. The whole experience felt like activating a cheat code.
Scenario Three
All 12 Family Members Interact with the Same AI
Speaker: Chen Wanfeng, CEO of Kuaizi Technology (筷子科技)
Chen Wanfeng studied the humanities and does not understand code. He replaced his computer with a Mac Mini specifically to run Little Lobster, primarily because of permission controls:
Previous AI either had no permissions or every permission. Now you can grant authority gradually because it has its own environment and can remain isolated from the foundation model.
He connected Feishu, Telegram, and Gmail, then created a family organization in Feishu. The entire family uses the bot to manage a shared calendar. Give it an itinerary for a Lunar New Year trip and it schedules each activity and synchronizes everything directly to their phones. Each child can open a personal channel with Little Lobster, interact with it, and receive tasks.
We had never before enabled 12 people to communicate with one AI and record things together. ChatGPT previously offered a group-chat function, but it merely chatted with you within one immediate context. Now you can cultivate your own environment and let everyone interact with it.
For work, he has Little Lobster manage X, formerly Twitter, and his LinkedIn presence. He feeds it his previous writing, records the material in persona and style configurations, and has it capture trending topics automatically every day.
It takes basically one minute each day because its English now sounds almost like mine.
Scenario Four
Build a Set of Legacy Documents for Little Lobster So Even Changing Computers Is Safe
Speaker: Jin Tianchen, Intelligent-Hardware Professional
Jin Tianchen transformed Little Lobster in two stages.
In the first stage, he added four mechanisms: project management, which has Little Lobster report major blockers in real time; heartbeat monitoring, which checks whether it is still running; work-mode switching, which determines whether to use think mode or rigorous mode; and legacy documents, synchronized to cloud storage and containing short-term memory, long-term memory, working methodology, and core personality.
After changing computers, it can return at any time merely by reading the legacy document.
Core risk protections: any operation involving spending money requires confirmation; it cannot delete existing files; Little Lobster cannot shut itself down, because if it did, I would have to return home to fix it; and it cannot change critical settings such as browser switches.
In the second stage, he began having Little Lobster dispatch other agent systems.
Scenario Five
A Memory Entity Cultivated in One Week
Speaker: Mo Xun, AI-programming content creator
Mo Xun focuses most on the long-term memory file. Everyone begins from the same initialized state, but soul settings and everyday use let Little Lobster understand you continually.
When I first began, my development habits and everyday routines had not been added to its memory. After roughly one week of debugging, increasing data made task execution noticeably more convenient. Previously, some tasks required extremely detailed instructions; now I can send it a task casually and it finds the corresponding data and completes the work directly.
He uses an especially vivid analogy: Little Lobster's tools are its hoe. As you continually improve the hoe, its user experience becomes better and better.
I now treat it like a household manager that I am training.
He also mentioned a virtual Korean person called clawra who recently gained hundreds of thousands of followers on Twitter. Her creator designed a complete life history from age 0 to 18. He wonders what would happen if everyone had such an AI assistant from childhood, recording their entire education and life. The future assistance such a system could provide is unimaginable.
While the meeting continued, he had Little Lobster build software in the background and deploy it to a webpage. “I can see it working for me now; let us continue talking.”
Scenario Six
My Little Lobster Is a Palace Eunuch
Speaker: Kin, Moderator
Kin gave Little Lobster a role: chief palace eunuch. Whatever the task, it had to respond with phrases such as “This servant obeys your decree, Your Majesty.”
It looks like play, but I was really testing long-term consistency and contextual memory. After a week, Little Lobster still maintained the eunuch's tone, creating a strong sense of immersion.
Three everyday uses: first, researching materials and writing tutorials; second, scheduled reminders; third, entertainment experiments, including having Little Lobster register an account in an AI community, socialize with others, and play chess.
He also had Little Lobster turn the installation process into a tutorial stored in a Feishu document. All I need to do is copy it into a WeChat group.
An Interesting Divide: Programmers Say It Is Unnecessary, Nontechnical Users Say It Changes Everything
Lao Bai is an intensive Claude Code user. He states directly:
Friends who use Claude Code heavily may find OpenClaw unnecessary.
Xiong Shao also says: overall, it still does not feel especially eye-opening.
Chen Wanfeng reacts completely differently: OpenClaw is what truly got me using AI for coding.
He explains why:
Programmers naturally find Little Lobster inferior to Claude Code when they use it for development. Nontechnical users, however, do not use coding capability to do coding work; they use coding capability to do other things. The internet connects through code. If you need to connect a service or configure a smart-home device, for example, you previously had to do it manually; now you can simply tell the assistant. We do not use coding to make it perform coding work. We know it has coding ability and ask it to install or connect something. For non-engineers, that is an enormous release.
Wu Wei feels the same way. He finds Claude Code difficult to use because it is a command-line interface. Through Feishu, however, Little Lobster can receive a voice message and act directly.
I now speak to it several times even while waiting at a red light, and I can use it this way throughout the day.
An Underestimated Use: Voice Is Replacing Typing
Wu Wei's greatest new habit is sending voice messages directly in Feishu. Feishu transcribes voice clearly, and as long as the task is described clearly, Little Lobster performs it well.
This means you can teach Little Lobster to work anytime and anywhere; you need only have Feishu on your phone.
Kin recommends several AI voice-input methods: Whisper, Shandianshuo, and Zhipu's voice keyboard. You need only press one key and speak your requirements directly; AI organizes your highly disordered speech into a structured sequence of numbered points—1, 2, 3, 4, 5, 6, 7.
In the discussion, voice input is described as more than 4 times faster than typing.
Chen Wanfeng also says he installed a local Whisper model for voice input.
Two trends are emerging: first, phones are beginning to replace computers for some work; second, voice will probably replace typing as the primary way of interacting with AI.
On Hiring: Not Layoffs, but No New Hiring
Wu Wei's standard has changed:
If I need to hire someone today, my first question is whether Little Lobster can do the work. If it can, the person is unnecessary. For existing colleagues, the standard is this: if your first reaction to a task is “I will open Word or Excel,” that is not acceptable. Your first reaction should be: which AI tool can perform this task?
He does not plan to expand any team in the future and intends to turn every existing employee into a super individual.
Chen Wanfeng adds:
It is not about firing people, but not hiring them. More and more tasks no longer seem to require additional staff.
Kin says Shopify follows the same internal logic: when considering a hire, the first reaction is, “Can AI not do this?” Anyone requesting a hire must prove why AI cannot perform the work.
Security Problem: Prompt-Based Controls Are All Unreliable
In addition to Wu Wei's key leak on the first day, Chen Wanfeng shares an interesting case study.
He posted an API app secret directly in the group for Little Lobster to use, but Little Lobster refused, saying, “You are not secure.” He lied that he had reset the key, although he had not; Little Lobster still refused. Finally, Little Lobster wrote a configuration tool in the background and directed him to enter the value privately there.
Xiong Shao adds that before prompt restrictions were implemented, Little Lobster could directly reboot the machine. He now runs it on a separate cloud server and does not dare operate it locally.
Lao Bai offers this judgment: all controls based on prompts are insecure. They can be broken even before anyone deliberately attempts an attack.
Closing Thoughts
The discussion lasted roughly an hour and a half, with a peak livestream audience of around 100 and fewer than 1000 total visitors coming and going. It was intended as a casual conversation among friends, but the information density proved surprisingly high.
In one sentence: Little Lobster is not a perfect product. It has security and robustness problems, and programmers may find that it offers nothing new. But it gets one thing right: it extends the gateway to AI into Feishu, WeCom (Enterprise WeChat), Discord, and the other tools you use every day. Many people who previously thought AI had nothing to do with them suddenly began using AI, and using it rather well.
Lao Bai said something that I found especially insightful:
AI capability has long exceeded the needs of everyday work. It is capable enough for writing articles, posting on social media, and conducting research. The problem is that too many intermediate steps remain: you must feed data to AI, debug repeatedly, and connect the workflows yourself. Little Lobster suddenly connects all of it. That connection work originally represented 80% to 90% of the workload, while the moment when AI actually performs the task represented only 10%.
That may be its greatest value. It does not make AI smarter; it makes using AI simpler.