"Old Boomer" Software Companies: What's Their AI Play? | Linear Voice
The AI Adoption Decision Matrix
Recently, legacy software giants in the US stock market have been facing a reckoning under the AI revolution. With technology iterating at breakneck speed, how should software companies build products in the AI era? This has become an unavoidable question for established players.
Today's share comes from Dr. Fan Ling, founder of Tezign — an angel-round portfolio company of Linear Capital.
He maps out a four-quadrant framework for how "last-generation software companies" can embrace AI, arguing that slapping AI features onto old software, while the most instinctive move for most companies, typically yields the same outcome: more features, messier interfaces, thinner margins. He also shares Tezign's own learnings from this journey, hoping it offers some inspiration.
Last time I was on the AI Alchemy podcast, the host actually called me an "old-timer" (lǎo dēng)! Fine, let's talk about what us "old-timer" software companies should actually do with product in the AI era.
I'll use a simple four-quadrant framework to describe the four choices:
- Old bottle, old wine: Keep the existing product unchanged
- Old bottle, new wine: Add AI features to the existing product
- New bottle, old wine: Rebuild existing demand to be AI-native
- New bottle, new wine: Use AI to address entirely new demand
Which is best? Honestly, you won't know until you try. But which is worst? I've thought about this for a while, and I have some answers.
#01
Why start by finding the "worst" strategy?
I emphasize "worst" not out of pessimism, but because with limited resources, eliminating obviously wrong directions is itself the most efficient strategy.
If you already know which path probably won't work, you can at least avoid wasting manpower, time, and organizational attention on it. Of the four strategies, the worst is: old bottle, new wine.
#02
Why is "old bottle, new wine" the most dangerous?
"Old bottle, new wine" means: layering AI features on top of existing software systems. This is almost certainly the most instinctive, and "safest-looking" reaction for any legacy tool company. But the problem lies precisely here:
First, enterprise software is already complex
Poor UX and high learning curves have long been the reality.
Second, adding AI often doesn't mean "getting smarter" — it means "getting more complex"
More features, messier interfaces, more confused users.
Third, the business results are usually worse
Users won't applaud; they'll just take it for granted. They won't pay more for these "add-on AI" features. Token consumption drags down gross margins.
The end result: the product doesn't become AI-native, but the organization gets consumed by AI first.
#03
What about the remaining three paths?
- Old bottle, old wine: Defending the status quo, nothing wrong with that
If a product already has clear PMF, stable users, and cash flow, maintaining it efficiently is itself a capability. This path isn't sexy, but it's rational.
- New bottle, new wine: Opening up a market that doesn't exist
Our own Atypica.ai, MuseAI, Clipo.cc are typical "new bottle, new wine": entirely new product forms, new technical paradigms, going after markets that weren't ours before. I've talked about this path quite a bit before, so I won't expand here.
- The truly hard one: new bottle, old wine
What I most want to talk about today is the third — "new bottle, old wine." This is the highest-difficulty strategy of the four. Why?
**"Old wine" means it already has PMF: existing users, value repeatedly validated. At the same time, it inevitably comes with massive R&D teams and technical debt: the legacy system needs continuous maintenance, refactoring costs are extremely high, and any big move risks cascading consequences everywhere.
So the question becomes: after you've already succeeded once, can you truly "start over"?
#04
Reassemble a team, start from zero
Our approach was to make a clean break, not upgrade the old system, but redefine the system itself: we formed a completely new, small core dev team focused on the "core problem" the original product solved, and wrote a truly AI-native product from scratch.
#05
What has content actually become in the AI era?
We've long worked on enterprise content management systems. Before large models, enterprise content mainly served marketing scenarios. After large models emerged, something changed fundamentally: content has become the "context" through which AI understands a company.
Because AI is naturally good at processing unstructured data, enterprise content is no longer just marketing material — it's the aggregation of corporate culture, decision logic, and knowledge structures. Its role has shifted from "production material" for marketing to "cognitive material" for large models. Based on this judgment, we built a new data asset management product: MuseDAM. We define it as: an AI-native data asset management system.
#06
From "human invocation" to "Agent invocation"
In the past, inside enterprises: content was mainly invoked by humans; or humans operated software to invoke it via API. But going forward, the invoking subject will fundamentally change: more and more invokers will shift from humans to AI Agents. This means:
- From "Human → API"
- To "Agent → MCP / tool chain"
This isn't just a technical change — it's a business model change.
I had a very visceral experience using Lovable: I don't understand databases, yet in the process of using it, I unconsciously paid quite a bit for its underlying database, Supabase.
The reason is simple — Vibe Coding Agent dramatically lowered the barrier to using databases. I wouldn't have been a paying database user before, but now I can use it through an agent. And because I don't understand it, I trust the agent's judgment more, and I'm more willing to pay.
#07
DAM is enterprise context for intelligent agents to invoke
Based on the same logic, I believe:
DAM (data asset management) will become one of the most important unstructured databases in enterprise AI applications, and the enterprise context that intelligent agents invoke.
In this system:
- It stores massive enterprise content
- Agent invocation results aren't just generating marketing materials
- They can also assist decisions, support innovation, derive next actions
Content is no longer just a marketing asset, but a company's most important "context asset."
#08
An example of "new bottle, old wine"
MuseDAM is our "new bottle, old wine" attempt. It continues the core value of DAM, but at the capability level, it's already an entirely different species. It's not "adding AI" to an old system, but rewriting the content system itself for AI:
- Works out of the box
- Natural language configuration
- Extremely friendly to AI Agents
- Fully adapted to AI's invocation patterns and usage habits
- Born as a global product
#09
This path must be taken
The difficulty of "new bottle, old wine" lies in:
- Having the courage to let go of existing entanglements and truly start fresh
- While not losing industry accumulation
But if you don't take this path, the old system only gets heavier. Patching and mending isn't easier — it will instead turn customers, experience, and historical accumulation from advantages into burdens that slow you down. This is a delicate balance between experience and reconstruction. Precisely because of this, it's a decision I can be proud of.
#10
Three directions Tezign is pushing forward
Next, we're focused on three things:
- Serving smaller-scale enterprises with more accessible AI-native systems: letting them have system capabilities on par with large enterprises
- Empowering existing clients to complete transformation with new AI-native systems: using the new data asset management system to help them truly enter the AI-native stage
- Serving international and China-outbound enterprises with globalized AI-native systems: using AI-native new products to replace expensive, outdated legacy systems overseas (like Adobe, etc.)
