Summary

AI can now handle collection, post-collection organization, processing and some coordination in a PPT workflow, but the finished product still hits a wall at aesthetics and accountability. The speaker compares himself to a chef: AI can help analyze ingredients, prep them and sequence the logic, but the human still chooses fonts, tunes color codes and adds labels page by page; he believes AI is already extremely strong at topic selection and judgment, but still cannot reliably deliver the finished work to his standard.
ima’s value goes beyond being “a slightly smarter bookmark folder”: it turns fragmented information into production inputs that can be retrieved on demand. The speaker spends 2–3 hours a day surfing the web and previously relied on WeChat groups to recover material; during May (month 5) alone, he accumulated 84 pages of visualized charts and 39 PDFs in ima—enough raw material for a PPT of a few dozen pages. AI’s organization and analysis of chart and report content is already quite usable.
The real need has moved from standalone tools to a collaborative assistant that runs across the entire knowledge-production chain. Side-by-side PDF translation addresses the difficult trade-off among being free, preserving the original layout and maintaining stable quality; Copilot, Agent, Skill and the future MCP are beginning to enter framework design and task orchestration. While preparing this presentation, the speaker already held multiple rounds of discussion with ima Copilot and adjusted the content accordingly.
AI-generated PPTs are now “good enough,” but still do not meet the delivery bar for high-standard content. The speaker’s 186-page 2025 AI industry review was first presented at an internal Tencent employee training session; at another internal Tencent training, he used 4 charts covering AI, social, gaming and UGC, with the first 3 made by hand and the fourth produced by NotebookLM. After extended use, NotebookLM has “almost only one use left” for him: turning a single very long article into a visual presentation.
ima’s next phase must confront questions around product form, monetization and its boundaries within Tencent’s ecosystem. The speaker cites a friend’s description of ima as an IDE for the knowledge domain, while noting that AI Agent tools appear to be abandoning the IDE; knowledge accounts and paid offerings in Coze’s redesign also raise questions around private-community management and how ima should differentiate from and connect with Knowledge Planet, Tencent Channels and WeChat. ima should consider more ways to generate revenue, as well as the platform’s own role in monetization and operations. Partnerships with Tencent Cloud WorkBuddy, Mavis and Yuanbao further blur the boundaries, rights and obligations between products; “These are all questions, and I don’t have the answers either.”
The so-called “last handcrafting” is not a rejection of automation, but a refusal to outsource the final judgment, aesthetics and responsibility for the work. 李世石 views Go as an art form and argues that “winning and losing are merely byproducts of the process of creating a perfect work”; the speaker’s conclusion is that as long as judgment, aesthetics and responsibility remain in human hands, “handcrafting will never truly end.”

Deep dive

1. “AI Sima Qian” Still Uses Marathon PPTs as the Industry’s Memory Bank

The speaker’s day job is strategy and investment at a Tencent-backed social and gaming company; he previously worked in VC, esports and livestreaming. His long-standing information-curation habit means his AI industry coverage spans industry, technology, capital, competition, fundraising, and both primary and secondary markets.
The production cadence is punishing: on April 26, he completed a roughly 98-page review of the AI industry from 2025 Q1 through the end of April; around early March, after the Lunar New Year, he produced a 70-page deep dive; and around January 20, he completed a 186-page 2025 AI industry annual review. On average, he produces a marathon PPT every 2–3 months.
Friends therefore call him “AI Sima Qian,” while podcast peers say he created the “PPT podcast” format. But he stresses that most of the charts come from reports, visualized materials and research institutions. His core contribution is not drawing charts, but processing the material, building the logic and deciding where each chart belongs in the story.

2. ima Turns WeChat-Group Hoarding into a Searchable Raw-Material Library

The speaker spends 2–3 hours a day browsing RSS feeds, WeChat public accounts, Substack, Twitter, Weibo and other information sources. He previously used his podcast fan group as a staging area: whenever he found a PDF, link or chart worth sharing or potentially useful for a future PPT, he dropped it into the group and later searched through the chat history to retrieve it.
WeChat groups suffer from a mismatch between control and context: they contain both his own material and other people’s messages and information, and were never designed for long-term knowledge production. ima provides a unified entry point, organizing content by month, topic, visualized charts and ordinary PDFs.
During May (month 5) alone, he saved around 84 pages of visualized charts in ima, averaging 2–3 pages a day. He also saved 39 PDFs, averaging slightly more than 1 per day. “The volume from this one month is basically enough material for a PPT of a few dozen pages.”
More important, AI’s organization and analysis of the content itself is already quite usable. Given a chart comparing the valuations and revenue of Chinese and US large-model companies, a brief analysis requires no fresh write-up from him. His warning is direct: “If you’re using ima only as a slightly smarter bookmark folder, I think you’re wasting it today.”

3. From Translation to Agents, AI Starts Taking on the Chef’s Coordination Work

ima initially solved a specific pain point. The speaker describes his English as poor, yet reads large volumes of English-language reports and charts every day. Side-by-side PDF translation addressed the trade-off among free use, layout preservation and consistent quality, and for a time became his most-used standalone feature.
Once he had dozens of charts, AI could also analyze what each one was saying and organize them into a report-style synthesis based on logic and framework. This was not direct slide generation, but the processing of raw material into “prepped ingredients” that could later be arranged.
His chef analogy defines the next stage: making a PPT is not just about stir-frying, braising or frying. It also means deciding how to handle the ingredients, what to do first and which tasks can run in parallel. “A chef’s job, fundamentally, is coordination—not chopping and dicing.”
That is why Copilot, Agent, Skill and the future MCP matter: they move ima from a repository toward task orchestration. While preparing this presentation, he held multiple rounds of discussion with ima Copilot about the occasion and the content, then revised the presentation accordingly—not merely saved and summarized material.

4. ima’s Real Test Is Product Form, Monetization and Its Boundaries Within Tencent

A friend once described ima as “an IDE for the knowledge domain.” But the speaker observes that the current generation of AI Agent tools appears to be abandoning the IDE, making ima’s product form an open strategic question.
Discussing Coze’s redesign, the speaker said Coze appears to have moved away from the IDE model and added knowledge accounts and paid offerings, naturally raising the question of private-community management. People in the relevant groups are also discussing whether more complex communication tools could be offered and how to integrate more closely with Tencent Channels and WeChat. ima needs to consider how to differentiate itself from existing private-domain tools such as Knowledge Planet while connecting with them where appropriate.
The speaker suggests that ima consider more ways to generate revenue, and argues that monetization and operations are not only the creator’s responsibility; the platform itself must also think through those questions.
Cooperation with AI Agent products across Tencent’s ecosystem creates another set of boundary questions. ima is already connected to Tencent Cloud WorkBuddy; Mavis is not yet connected, while Yuanbao is expected to connect soon. QQ Browser, Tencent Docs and Tencent Meeting have similar AI features. As product boundaries blur, ima and these “family members” will need to rethink their respective boundaries, rights and obligations. “These are all questions, and I don’t have the answers either.”

5. AI Can Generate the Last Mile, but the Meaning of the Work Still Cannot Be Outsourced

The speaker does not deny that AI PPTs are already usable. He can feed dozens of charts into ima and receive an organized output; where the quality bar is not especially high, AI is capable of completing the final step.
His 186-page 2025 AI industry annual review was first presented at an internal Tencent employee training session. At another internal Tencent training, he covered AI alongside social, gaming and UGC, using 4 charts in total: the first 3 were made by hand, while the fourth was produced by NotebookLM.
After extended use, NotebookLM has “almost only one use left” for him: turning a single very long article—especially one in English—into a PPT-style visual presentation. His NotebookLM library consists almost entirely of single-source material rather than multi-source synthesis. Nano Banana, in his view, has a step-change lead in visual output, particularly images, and performs extremely well, but seems to have ended up as a kind of translation tool.
The speaker believes AI is already “very, very strong” at topic selection and judgment, but still does not appear to have replaced humans when it comes to aesthetics and responsibility. He relayed a line from a more than 40,000-character article by the newly installed pope: “Everything can be outsourced, but love cannot be outsourced, responsibility cannot be outsourced, judgment cannot be outsourced, and meaning cannot be outsourced.”
李世石’s experience makes the point concrete. At the end of April 2026, 10 years after AlphaGo’s victory, he appeared on a Korean program with Faker and said that, from the human perspective, Go still has infinitely many solutions, so every player must create a one-of-a-kind style. From the perspective of art, he said, “Winning and losing are merely byproducts of the process of creating a perfect work.”
AlphaGo defeated 李世石 in 2016, and he retired 3 years later. The program’s host said the public had lost a Go artist too soon and called him “the last guardian of Go as an art form.” The speaker turns “the last handcrafting” from a mode of operation into a creative stance: “As long as we are willing to keep judgment, aesthetics and responsibility in our own hands, handcrafting will never truly end.”

May 2026 Random Ramblings Ivanka Trump on Building an Authentic Life