張志祺：重塑組織導入AI的體驗設計

1. In one sentence

Simpleinfo co-founder 張志祺 (of 志祺七七) described how his team went from last year's methodology of decomposing flows, chaining tasks and building shared spaces to realising that what actually blocks organisational AI adoption is a psychological barrier rather than a technical one — and so turned toward "setting the technology aside and using experience design to fold AI painlessly into the existing workflow", illustrated by a Slack bot triggered by an emoji reaction and a Google Sheets add-on.

2. The session

The speaker's background

張志祺 is a co-founder of Simpleinfo (brand name 圖文不符) and the host of the YouTube channel 志祺七七. Simpleinfo was founded in 2015 by 張志祺 and 王成祥, arguing for folding information design into aesthetics and communicating issues to the public through explainer videos and similar formats (verified against Wikipedia).

The host introduced 志祺七七 on stage as "a YouTuber with 1.6 million subscribers"; on checking the channel's current count, Wikipedia shows it has reached 1.68 million (the channel passed a million subscribers on 5 June 2022 and received a YouTube gold award, with over 1.8 billion cumulative views). The two figures do not conflict; the verification simply happened later than the stage introduction, and the body of these notes uses the newer 1.68 million.

The three core practices he shared last year (2024)

He first ran quickly back over what he presented on the same stage last year, when he offered three practices for helping a team adopt AI: first, decompose the flow — break the whole task into details and match different AI tools to different steps rather than expecting one model to solve everything; second, chain the tasks — use a flow board to link tools into an automated pipeline so AI plays an assistant's role inside the flow; third, build a shared space so that people who can and cannot use AI can both take part, understand the design logic, and find their own role and sense of achievement.

This year's gap: better tools, lower usage

A year on, AI tool quality has visibly improved and the barrier to writing prompts has all but vanished — and yet the team's actual frequency of using AI has fallen. He used the Instagram sponsored shoot flow he shared last year as an example: understand the client's need, generate characters, design the story, plan the storyboard, assemble the shoot plan — each step with a matching AI prompt card (character generation card, story generation card, storyboard generation card) with prompts and sample outputs behind them, forming a whole flow board that runs the process end to end. But even with the flow laid out this completely, people still got stuck at the first step. Digging in, the sticking point was not that the flow was unclear but that the psychological barrier of "changing the way you're used to working" is harder to cross than the technical one.

He breaks that psychological barrier into three layers: first, most people are not good at decomposing a task into steps; second, even if you can decompose it, you may not realise "this part could be handed to AI to collaborate on", which is a new working mindset and judgement that takes practice; third, even with that awareness you still have to know how to turn that stretch of flow into a prompt an AI can understand and complete effectively. He quoted something a doctor said to him last year when he sought help under stress: "humans are energy-saving creatures", habitually taking the easy path — one more step, one more unfamiliar screen, and we instinctively avoid it, much as adding a single SMS verification step at online checkout makes many people simply abandon the purchase.

The turn: make AI fit people's habits rather than asking people to learn AI

That observation made them invert the whole logic: rather than requiring everyone to learn to use AI and proactively apply it at work, make AI appear more naturally inside the existing flow, so that with no extra learning and no new operation to adapt to, AI becomes part of daily work.

Case one: the "伸手牌" Slack bot

The situation: the team receives a large volume of sponsorship enquiries every day, five or more on a busy day. The old flow had the account manager drop the client's requirement into a Slack channel, and the planner would then open a flow board or a ChatGPT window, copy the requirement across manually, generate several angles, filter and organise them, and paste them back into Slack for discussion — a back and forth that constantly interrupted the planner's train of thought and often meant three to five hours or even the next day before replying, by which time the opportunity had frequently gone to a competitor.

The solution was a tailor-made AI feature inside Slack: react to the message thread carrying the client's requirement with a "reaching out" emoji, and it triggers a small AI bot they call "伸手牌", which follows a pre-set prompt to reply automatically with a script suggestion and several angles. The planner no longer starts from nothing but "starts at eighty out of a hundred", concentrating only on judging which angles are viable, with final quality control still resting with a human. He thinks the case's biggest highlight is that every action stays inside the Slack interface people already know — a new joiner needs no new tool and no prompting skill, and an action as natural as one emoji triggers AI collaboration.

Case two: a Google Sheets add-on generating caption cards

Another common pain point is captioning videos: the planner used to copy long passages from a spreadsheet into ChatGPT one section at a time and paste each result back, and switching windows repeatedly made it easy to paste in the wrong place or miss content — cumulatively a real drag on focus. Their solution was a Google add-on integrated directly into the spreadsheet: the planner clicks a button and the system automatically numbers each section (making it easy to identify which section to reshoot while filming, and to annotate feedback while editing), and clicking again has AI generate the caption text for the corresponding column by number, so the planner only needs one final check before handing it to the editor — no more switching tools, no more manual copying and pasting.

He also shared what he imagines next: with more automation chained on, perhaps combined with the new image generation techniques recently released, maybe the matching visual assets could be generated and filed automatically too; and together with the automated editing bot they have already built, which cut rough-cut time from an hour and a half to about fifteen minutes, he thinks the time to complete a sponsored short could drop within the next year from two to two and a half hours to five to ten minutes. All of these time figures are his own expectations of the future, with no external corroboration.

The common pattern behind these cases

He distilled a very simple process: it usually starts with somebody on the team thinking "if there were a magic button here to automate this, it would save a lot of trouble", and proactively going to the internal AI team to discuss feasibility; the AI team takes it from there, clarifying the requirement, decomposing the flow and building a prototype to validate whether AI genuinely helps. Nothing in the process is technically flashy; what matters is repeated discussion and fine-tuning. He considers this a question of culture and organisational structure: the front line has to feel psychologically safe enough to voice questions and frustrations about the flow, and the organisation has to have people ready to catch those questions and provide support, so there is no barrier between "raising a problem" and "getting help".

Replicating it externally, and concrete advice

This experience design methodology is no longer only internal; it has helped other organisations — he named Business Weekly and TaiwanPlus as organisations that came to them, because their managers had observed the problem of "a pile of AI tools that nobody uses" in their own teams (speaker's own account; no public material found corroborating the details of collaboration with Business Weekly or TaiwanPlus). The principle is the same: change the other party's existing flow and behaviour as little as possible, and have AI show up proactively in the right place to support decision and execution.

His advice for organisations wanting to adopt AI is concrete: go back and examine your own workflow for the steps where execution stalls, judgement slows, information is unclear between people, or colleagues keep asking the same question — those are the candidate points for AI. He also warned that this is genuinely time-consuming and difficult: past a certain size, the people doing strategic thinking and the people doing front-line execution drift further apart, and people who can see both the sticking points and the technology are scarce. He added one crucial condition — the volume has to be large enough to justify automating: plenty of YouTuber friends who publish one episode a week have no incentive to bring AI tools into their flow because the volume is not there, whereas Simpleinfo produces one main-channel video a day plus five to ten Shorts, at a scale that makes the investment worth it. He noted this echoes Jensen Huang's earlier remark that technology renewal requires scale.

He also drew an honest line between doing it yourself and getting outside help: a planner teaching themselves prompting or wiring up a simple tool sounds like something a little research can solve, but the genuinely complex problems — data security, version maintenance, fixing errors arising from tweaked requirements — only surface when you try to share a tool with the whole team, and those are no longer solvable by self-teaching or prompting. They are the professional territory of automated flow design and operations. His company now offers that kind of contract work, describing it as "something a project might sort out for a few tens of thousands", but he still recommends enterprises build the corresponding capability internally.

The biggest change of the year

He compressed the year's change into three contrasts: from "focusing on improving individuals' ability to use AI" to "starting from the design of interface and experience"; from "expecting users to actively direct AI" to "having AI actively intervene at the right point in the work"; from "insisting on making AI tools better" to "caring about folding AI painlessly into the existing workflow". He believes what drives an organisation's AI adoption is not technical innovation but experience design — better than making AI smarter is putting it in the right place, so that it appears at the right time, in the right interface, in the right language. That is where reducing friction genuinely begins.

3. Figures and cases

4. Lines worth keeping

• "Humans are energy-saving creatures — one more step, one more screen, one more unfamiliar operation, and people basically avoid it instinctively."
• "Real automation isn't just doing a few steps fewer; it's making people barely notice they've changed anything."
• "What genuinely drives an organisation's AI adoption is not technology but experience design."

5. Tools and terms mentioned

• Simpleinfo, 志祺七七: the speaker's company and personal brand
• "伸手牌": the internally built Slack AI bot triggered by an emoji reaction
• A Google Sheets add-on: the AI tool integrated into the spreadsheet that generates caption cards in one click
• The automated editing bot: the internal tool that sharply cut rough-cut time
• Flow boards: how last year's methodology chained AI prompts into an automated flow
• Character generation card, story generation card, storyboard generation card: prompt cards on the flow board matching each step

6. Wider observations

The most instructive line was "humans are energy-saving creatures" — translating "people don't want to use AI" into "one more unfamiliar screen and people give up" is a far more honest acknowledgement of the cost of changing working habits than blaming staff for not learning. It is very different from a purely technical talk: he barely mentions a model or a parameter, spending the whole session on interfaces and trigger points — and that may be closer to where most small and medium teams actually get stuck than any technical architecture.

His threshold of "the volume has to be big enough to justify automating" also partly explains why the E.SUN Bank case 黃仕鎮 shared the same day was worked out in such detail — a bank's headcount and transaction volume are inherently large enough to justify investing in a whole governance and automation apparatus with a high setup cost, which is the same logic as 志祺七七's one main video a day plus five to ten Shorts, at a different scale.

7. Sources