This is a republication of Reedo’s Naver article. It is not a new experiment or a DMS client case study. The cover and in-article photographs are AI-generated staged scenes, not evidence of actual work.
Even after installing an AI browser, you may end up using it like a search box if you have not decided what to delegate first. This video presents seven steps, starting with a small task that asks Aside to read a website and create a table, then expanding to recurring tasks and instructions sent from a phone. The key is to define both the results and the boundaries, rather than granting every permission from the outset.
The directly demonstrated comparison of pricing claims, channel-metric reading, and accommodation comparison should be distinguished from external reference examples. In particular, the video explains that the values in the collected tables are what the agent read and were not independently cross-checked. A quickly assembled table should therefore not immediately be treated as confirmed information.
Verify one read-only task, then add memory, scheduling, and remote instructions in that order.
AI-generated staged image, not an actual work screen.
How should you choose your first task?
The first step is to install Aside through the official channel and connect the model you want to use in settings. The video shows the model-selection screen for different kinds of work and explains that the model for a scheduled task was changed in the routine editor, not in the conversation. Even when using the same AI, you should not assume that the current conversation and scheduled runs share the same settings.
The second step is to specify the website, the desired result, and when to stop. For example, set the scope to reading public pages and organizing items into a table, without changing or sending anything. The quick table creation in the video is the result of that particular request, not a guarantee of the same completion time on every website. Ask for unknown values to remain blank or be marked as unverified, so the starting point for comparison is clear.
AI-generated staged scene 2View original
AI-generated staged image, not an actual work screen.
Where should permissions stop?
The third step is managing permissions and passwords. The video recommends starting in read-only or Guard mode, then expanding access only when writing is necessary. Final actions with external effects, such as sending, paying, deleting, or submitting, should remain checkpoints for a person. Even when granting access to read inside an account, first check that the channel and date range match what you intended.
In the actual channel-metrics demonstration, a different channel was initially open, so the channel was switched and the request made again. A table of views, watch time, and other metrics was produced, but those figures were not separately cross-checked. The accommodation comparison also involved different room types and consecutive-night discount conditions, so the lowest price alone was not enough to decide what to book. Checking the original pages remains part of the reader’s work.
The fourth step, memory and skills, helps reduce repeated explanations. The suggested sequence is to distinguish memories accumulated through conversation from task-rule files, try delegating the task two or three times to find an approach that works, and then formalize it as rules. It is safer not to expect memories from other AI services to merge automatically.
AI-generated staged scene 3View original
AI-generated staged image, not an actual work screen.
When should you add scheduling and phone instructions?
The fifth step, routines, is to begin by scheduling one task that has worked successfully. The video describes an attempt to save a routine with higher permissions than the conversation allowed; it was blocked and then saved again with Guard permissions. It also states that both the computer and Aside must be running, and warns against assuming that missed work will automatically be caught up later. Do not just create a schedule; check the actual results over several days.
The sixth step connects Aside to coding tools through the command tool in developer settings and MCP, and the seventh uses a connected messenger to send instructions from a phone. The video confirms Telegram replies and receipt of scheduled reports, but also shows a remote request whose subtask could not open the browser and therefore did not reach publication. Since the cause was not established, it is better to test important tasks at the computer first and begin remote use with instructions and status checks.
Watch this video to see the actual process from installation to remote instructions
Today’s starting point is not complicated automation, but a read-only summary of one public page. Include the source, output format, and stopping conditions in your request, and personally cross-check the table’s key values. Once that task is reliable, adding skills and routines can help you define not just what is convenient, but how far you can trust it to work on your behalf.
