Des Traynor - Real Talk about AI and Software - [Invest Like the Best, EP.340]

Here is the English translation of the provided Chinese investment research notes, following all specified rules.

At a Glance

Des Traynor (Co-founder & Chief Strategy Officer of Intercom) shares how AI is practically transforming his customer service business, along with his thoughts as an investor. The core theme is: AI’s value in empowering existing businesses is far greater than its potential to disrupt startups. The key lies in deeply integrating AI into existing workflows, rather than merely adding an "AI feature." The most impactful judgment of the entire piece: Des Traynor believes the software industry suffers from "unsustainable fragility" — code is easily copied, product value depreciates rapidly, and adding complexity to pursue market breadth inevitably makes the product worse, a stark contrast to traditional consumer goods like Coca-Cola.

Theme 1: AI's Advantage for Incumbents Far Outweighs That for Startups

Des Traynor argues that for most software categories, incumbents are better positioned to benefit from AI than startups. He proposes a core analytical framework: the key to determining whether an AI startup has a chance is whether "the new architecture fundamentally overturns the core assumptions of the existing product."

• Mechanism Breakdown: Traynor uses MailChimp as an example. Suppose an AI startup wants to challenge it. Even if the AI can automatically generate email content and designs, the startup would still need to build the massive email delivery platform, link attribution, cross-client rendering tests, and other infrastructure from scratch. Meanwhile, MailChimp only needs to call OpenAI's API to enhance its existing features. In this scenario, the incumbent (MailChimp) only needs to do 20% "new work," while the startup needs to do 80% "old work."
• Data Chain & Deduction: He further explains with quantitative thinking: "Suppose your new startup and mine can move 10 times faster than MailChimp. We finish all the AI features in 3 months, while MailChimp needs 30 months. The question is, in the remaining 27 months, can we match MailChimp's standards on all other features?" If the answer is no, the startup has no chance.
• Counterexample: Traynor also identifies areas where startups can win. For example, an AI-driven ad management optimization tool that can automatically create, test, and adjust ads, completely replacing manual operations. In this scenario, the core assumptions of existing ad platforms (like Google Ads) — dashboards, reports, configuration processes — are completely overturned, giving the startup a massive advantage.

Theme 2: Fin's Success Comes from "Workflow Deletion," Not "Feature Addition"

Des Traynor emphasizes that the success of Intercom's AI bot, Fin, stems not from being a better "copilot," but from directly "deleting" a large number of workflows in customer service. He advises other companies to focus on "which tasks can be completely removed" rather than "adding AI assistance to existing tasks."

• Data Chain: After Fin's launch, customers saw an average immediate reduction of 15-25% in support volume, with some customers experiencing a 50% drop. Fin is priced at $0.99 per resolution. Traynor believes this is cheaper than most human agents (especially in B2B tech companies) and provides the instant value of "zero-second response."
• Mechanism Breakdown: Fin's core is not retraining a model, but using carefully designed "vector search" and "prompt engineering" to constrain OpenAI's general capabilities within a customer's specific knowledge base (help docs, historical conversations, etc.). The key is: Make the AI only answer questions relevant to the customer's business, refuse irrelevant queries (e.g., "Who is the US President?"), and reduce "hallucinations" by limiting the information source.
• Deduction: Traynor believes the ultimate future of customer service is a flywheel of "humans helping bots, bots helping humans." A human agent only needs to handle a new problem once; the AI learns from it and generates a "knowledge snippet," ensuring the problem doesn't recur. Human time will shift towards high-value brand building and proactive support.

Theme 3: The "Curse" of the Software Industry — Replicability, Rapid Depreciation, and the Complexity Trap

Des Traynor presents a sharp perspective on the nature of the software industry: software is not a perfect business model; it has three inherent, difficult-to-solve weaknesses. This view forms the core of his investment and entrepreneurship philosophy.

• Point 1: Lack of a Durable Moat. Software code and UI are extremely easy to copy. "All SaaS is essentially a UI on top of a database," so "anything you have is almost destined to be copied." He argues the only moat is "product momentum" (continuous rapid iteration), which makes it daunting for latecomers.
• Point 2: Rapid Value Depreciation. Software products have a very short "shelf life." "The best product in the project management space three years ago might be far from good enough today." Companies must continuously invest heavily to maintain their position, a stark contrast to products like Coca-Cola.
• Point 3: The Contradiction Between Market Breadth and Product Simplicity. To serve a larger market, software must add more features, settings, and permissions to accommodate different customer workflows. This makes the product increasingly complex, degrading the experience for any single user. Traynor concludes: "For a product to become big and successful, it must get worse." He believes software like iOS Notes or Craigslist — "one product serving everyone, with everyone using it the same way" — is the ideal state, but extremely rare.

Theme 4: Evolution as an Investor — From "Product Aesthetics" to "Business Fundamentals"

Des Traynor shares his evolution as an angel investor, moving from being captivated by "beautiful products" to focusing on more fundamental business questions.

• Early Mistakes: He was once seduced by "products that looked great," believing "making good software was enough."
• Current Framework: He now uses a "four-element" question to screen startups: "Can you say one thing that is unique to you, valuable to users, true, and simple?" He believes most companies fail on at least one of these four elements.
• Key Insight: He gives the example of a failed AI meeting note-taking tool. The product itself was "exceptionally good," but the problem was "nobody wanted to pay for it on top of Zoom and G Suite." This made him realize that "the go-to-market path multiplied by the user's willingness to pay is a real and brutal constraint."
• Sharp Advice: He bluntly criticizes startups that achieve "74 months of runway" by cutting costs, calling it equivalent to "wasting the best six years of your career." He advises founders to set a 6-12 month growth deadline and end it decisively if not met, because "the worst outcome isn't the company failing; it's you wasting your life."

Position Moves

Target	Guest's Stance	Key Data
Intercom	Bullish (as own business)	Serves 25,000 businesses; Fin reduces support volume by 15-50%; Fin priced at $0.99/resolution; Handles 500-600 million conversations monthly.
OpenAI	Bullish (as tech supplier)	GPT-4 is "far ahead" of other models in capability; its pricing (GPT-4 is expensive) is the main source of Fin's cost.
Anthropic	Neutral (potential alternative supplier)	Could be an alternative due to regional availability (e.g., EU AWS instances) or pricing.
Rewind AI	Bullish (as investment case)	Believes its model of "ingesting massive data and providing a natural language query engine" is very powerful.
Kittle	Bullish (as investment case)	Generates vector graphics from descriptions, allowing subsequent user editing; a good example in the AI visual space.
Equals	Bullish (as product case)	Its AI feature allows users to write Excel/SQL queries in natural language, significantly lowering the barrier to entry and expanding TAM.
MailChimp	Neutral (as analysis case)	Used to argue that incumbents (needing 20% new work) have an advantage over startups (needing 80% old work).
Linear	Bullish (as competitive benchmark)	Seen as a benchmark for "hard-to-copy" project management tools; its product momentum itself is a moat.
Stripe	Bullish (as competitive benchmark)	Also seen as a benchmark company with strong product momentum and brand position.
Zoom	Neutral (as analysis case)	Its strong market position and bundling with G Suite make it difficult for startups building paid AI features (like meeting notes) on top of it to succeed.

Judgments Worth Remembering

1. The key to judging whether incumbents or startups have the advantage is whether "the new architecture overturns the core assumptions of the old product." If 80% of the old infrastructure is still needed, and the incumbent only needs to call an API, the startup has no chance.

2. "Don't build another 'copilot'; instead, think about 'which tasks can be completely deleted.'" The real value of AI lies in end-to-end automation, not adding an AI assist button to an existing process.

3. "All SaaS is essentially a UI on top of a database, so everything is replicable." Software lacks a durable moat; the only defense is "product momentum" — running fast enough to leave latecomers in the dust.

4. "For a product to become big and successful, it must get worse." To serve a broader market, software must add complexity, which inevitably harms the experience for any single user. This is an inherent contradiction in the software business model.

5. "The value of software is highly perishable. The best product three years ago might be far from good enough today." Software companies must continuously invest heavily to maintain their position, a stark contrast to traditional consumer goods like Coca-Cola.

6. "The go-to-market path multiplied by the user's willingness to pay is a real and brutal constraint." No matter how good a product is, if users lack the willingness or path to pay for it, it cannot succeed.

7. "The worst outcome isn't the company failing; it's you wasting the most valuable years of your life pushing a rock up a hill." He advises founders to set a 6-12 month growth deadline and end it decisively if not met, rather than extending a "zombie period" by cutting costs.

8. "The inherent risk of AI models not being 100% accurate is real, especially in high-stakes domains." He suggests that the ease of use of AI can actually introduce a risk of "reduced attention" (like assisted driving), where users may over-trust the system and fail to recognize its errors.