Leo Jiang spent his career at the intersection of large enterprises and the outside world: BT, Orange, Vodafone — senior positions across Asia, always in consulting and innovation. Then he joined Huawei.

The reason was partly professional (“I’d never worked at a Chinese private company”), partly timing (“Huawei Cloud was a massive business group, $30–50 billion scale”), and partly — as he puts it directly — a touch of patriotism. Huawei was under sustained US pressure. He felt he should do something.

What he built there is what makes this conversation worth reading for anyone thinking about startup ecosystems, AI go-to-market, or the future of B2B sales.

Building a Startup Ecosystem Inside a $130 Billion Company

Leo’s mandate at Huawei was growth through the startup ecosystem. He started the Spark Program — Huawei’s first overseas startup accelerator — in August 2020. The first cohort received 450 applications. By year four: 7,000–8,000 applications spanning all five continents.

The growth wasn’t accidental. Leo describes two parallel challenges: externally, understanding what founders actually need and delivering it; internally, convincing a $130 billion company that startup ecosystem value doesn’t show up in quarterly metrics.

“Everyone wanted us to find the next Grab or Netflix for Huawei to ride with,” he says. “That’s not how ecosystems work. You can’t even reliably pick winners by investing in them — let alone by running an accelerator.”

What he did differently was give founders center stage. The Spark Founder Summit — launched during COVID in 2021 — put startup founders at the center, with Singapore government ministers, Hong Kong officials, and unicorn founders (including Klook, WeLab, and Carousell leadership) in the audience to listen, not speak.

“To this day, I haven’t seen any other company — Western or Eastern — genuinely build a stage for founders. Not a panel where they get seven minutes. A stage that says: the government, the VCs, the unicorn founders are here to listen to you.”

Why He Left: Four Years of Watering the Seed

The four years at Huawei changed Leo in a way he didn’t expect. Daily contact with founders — running alongside them, hearing their problems, celebrating their wins — slowly grew his own desire to build something.

But he’s honest about the gap between advising and doing: “I realized I was being hypocritical. I was giving advice from a position I’d never been in. My perspective was probably useful because it was different, but I couldn’t actually empathize with what founders were going through.”

He describes the Huawei years as “watering the seed.” The moment it broke through the soil was when he saw the AI opportunity and recognized that his strength — understanding how businesses grow across markets — mapped directly onto an underserved problem.

The Case for AI Sales Agents

Leo’s company, The Grow, builds AI agents for outbound B2B sales. To understand why this matters, you need to understand the outbound sales workflow.

In B2B, there are roughly 100 million people doing sales globally, and 40–45 million of them are doing outbound — the proactive side of sales where you find potential customers, research them, reach out, build trust, and book meetings. This work is structured into six or seven discrete steps, each with its own specialized tools and providers.

The people doing this front-line work — SDRs (Sales Development Representatives) — typically last less than two years. The work is brutal: sending hundreds of emails, making dozens of calls, facing constant rejection, all measured by a single metric: meetings booked per week. “It’s an extremely draining job. No human wants to do it,” Leo says. “This is exactly the kind of work AI should do.”

The Grow automates the entire outbound pipeline with four specialized agents: a search agent (finding the right companies and contacts), an analysis agent (scoring whether they’re worth pursuing), an outreach agent (drafting and sending personalized messages), and an optimization agent (improving the whole system over time).

Autonomous Sales: L1 to L5

The most interesting design decision is The Grow’s autonomy framework, explicitly modeled on self-driving car levels.

At Level 1–2, the system does research and analysis. You tell it “I want to sell to DBS Bank” and it maps the organization, identifies the right contacts, analyzes recent company developments, and presents a brief. But it doesn’t act.

At Level 3, it drafts outreach campaigns — complete emails, contact sequences — but waits for your approval before sending anything.

Level 4 sends autonomously, but escalates edge cases: ambiguous responses, time-sensitive situations, anything it can’t confidently handle. This is the “human in the loop” level.

Level 5 is full autonomy — the agent books meetings directly on your Google Calendar and notifies you via Slack.

The most remarkable capability, according to Leo, isn’t any single step — it’s what the combined system can do that no human could. He describes a recent case: a US-listed client was attending a trade show in Japan. They fed The Grow the client’s website and the exhibition website. The agent identified all exhibitors and keynote speakers, analyzed which ones were potential customers, ranked them into three tiers with detailed reasoning, and produced a targeting strategy — work that would have required days of manual research plus specialized web scraping tools.

“This is the thing that excites me,” Leo says. “The agent can do things a human salesperson literally cannot do.”

SaaS Is Dead. What Comes Next?

Leo’s view on the software industry’s future is unambiguous: “SaaS as we know it is ending. I’m not saying SaaS as a category will die, but the form will change completely.”

His reasoning: when AI models can handle complex multi-step workflows end to end, the value of a human operating a software interface drops toward zero. The Grow’s own evolution illustrates this — version 1.0 was automation tooling, version 2.0 was a SaaS product focused on search, but version 3.0 is an autonomous agent. Users don’t operate it; they instruct it.

The pricing model follows the logic: The Grow has shifted entirely to outcome-based pricing. No seat-based fees. If the agent delivers meetings, you pay for meetings. If it doesn’t, you don’t.

“If you’re building a sales agent — or any agent, really — and you’re still charging per seat, you haven’t understood what you’re building,” Leo says. “You should be charging like you’d pay a salesperson: a base cost for doing the work, plus commission on results.”

Agent vs. Agent: The Inevitable Future

The conversation’s most provocative moment comes when Leo describes what happens when AI sales agents become widespread.

“Right now, if you use a sales agent and your competitor doesn’t, you have a massive advantage. But what happens when everyone has one? Every company is sending agents to sell. Customers get overwhelmed. So they deploy their own agents — procurement agents — to handle the incoming flood. It becomes agent-to-agent. This isn’t hypothetical. It’s happening right now.”

He references Anthropic’s March 2025 report on AI’s impact on the job market, which shows a wide gap between AI’s theoretical capability to replace jobs and its current practical capability. For sales, the theoretical ceiling is about 70% replacement, but the current state is much lower. “Whoever closes that gap fastest wins,” he says.

When asked whether anyone can just build their own sales agent, Leo is direct: “Mathematically, LLM hallucination cannot be 100% solved. That gap means full replacement is impossible. But it also means whoever gets closest to closing it captures the most value.”

What Chinese Founders Get Wrong About Southeast Asia

Leo has a sharp view on Chinese AI companies entering Southeast Asian markets — informed by years of watching both sides at Huawei.

“Chinese companies come in with three moves: our technology is strong, we have big domestic clients, and our price is incredibly cheap. That doesn’t work here.” He notes that Singapore and Southeast Asia’s enterprise market, while smaller than North America, tends to follow Western purchasing patterns. Trust-building and communication style matter far more than price.

“The paying willingness and enterprise density in Asia is lower than North America. That’s just reality. But there is an advantage: the market is slightly behind on adoption, so if you have something genuinely new, there’s a window.”

The Grow’s own client base reflects this: a mix of Chinese companies going overseas, Southeast Asian startups, and Western enterprises — each requiring different approaches, none satisfied by the “China playbook” alone.

This article is adapted from 离线时间 EP09.