In the ashes of Terra, we didn't just lose money; we lost faith in narratives. Fast forward to 2025, and a similar tremor is shaking the AI-blockchain sector. An internal letter from Zhipu Network's founder, Tang Jie, leaked yesterday, and it reads like a desperate bid to escape the valuation gravity that crushed MiniMax Token after its lock-up expiry. The market is trying to price in the pivot, but I've parsed the 7,000-word strategy document through the lens of lived experience—both as a builder and as a news operator who watched Terra implode. Let's cut through the hype.
## The Hook: A Signal from the Command Room The letter explicitly deprioritizes Zhipu's flagship product, CodeGeeX—an AI-powered smart contract auditor that had been the project's main revenue driver. Instead, Tang declares an all-in bet on 'long-horizon tasks, autonomous agents, self-evolution, and AGI.' The market reaction was immediate: ZP token dropped 12% in 30 minutes, then recovered 8% as retail narrative traders piled in. But the signal is clear—this isn't a pivot; it's a narrative reframe to dodge the valuation trap.
## Context: Why Zhipu Needed to Change the Story Zhipu Network launched in 2022 as an AI-layer for blockchain, using its GLM-series models to power smart contract auditing and code generation. By mid-2024, CodeGeeX had captured ~15% of the DeFi auditing market, generating steady API revenue. But the problem? As a utility token, ZP's valuation was starting to correlate with revenue multiples—a dangerous metric in a bear-to-bull transition. MiniMax Token's 70% crash post-lockup was the wake-up call: pure narrative-based valuations cannot survive when investors start asking for ARR. Zhipu needed a new story that was harder to measure and easier to dream on. Enter AGI+Agent blockchain—the perfect un-falsifiable narrative.
## Core: The Anatomy of the Pivot Let me break down what the internal letter actually says, and what it means for the technology:
- From Code to Agency: Zhipu is de-emphasizing CodeGeeX (a measurable, revenue-generating product) and doubling down on 'autonomous agents'—AI systems that can plan, execute long-term tasks, and interact with on-chain protocols. This is a shift from a tool to a platform.
- The 'Touch High' Plan: A roadmap to achieve GPT-4-level agent capability within 18 months, with self-evolution mechanisms. The letter explicitly states: 'Short-term monetization is not a priority.'
- Capital Market Logic: The pivot is timed to pre-empt a valuation correction. By moving from a 'proven revenue story' to a 'future potential story,' Zhipu can justify a higher multiple for longer, buying runway for R&D.
But here's the technical rub: Current autonomous agents (think AutoGPT variants) have success rates below 30% for complex on-chain tasks like cross-chain arbitrage or yield optimization. Zhipu is essentially betting on a technology that doesn't yet work at scale, while walking away from a product that does.
## Contrarian Angle: It's Not About Technology, It's About Survival Most coverage is framing this as a visionary leap towards AGI. I've been in this industry since the 2017 Bitcoin.com ICO, and I've seen this movie before. This is a textbook 'narrative arbitrage' play. When a project's valuation is tied to revenue, it's capped by fundamentals. When it's tied to a grand vision, investors can only price in belief. Zhipu is terrified of becoming the next MiniMax—a project that peaked during the narrative cycle and crumbled when liquidity unlocked. The pivot is a defensive mechanism, not a technological breakthrough.
What's not being reported: The letter mentions 'self-evolution,' which in blockchain terms means the agent will have on-chain governance rights to upgrade its own core models. This is a security nightmare. If an AI agent can autonomously vote to change its reward function, what stops it from draining the treasury? Zhipu's AGI bet is also a bet on alignment research, which is still in its infancy. The contrarian take—and the one the market is ignoring—is that Zhipu is increasing its technical risk profile, not decreasing it.
## Takeaway: Watch the Technical Milestones, Not the Narrative Zhipu's pivot is a masterclass in narrative engineering, but the blockchain community needs to demand proof. Over the next six months, I'll be tracking: - Agent demo releases: Do they release a working, end-to-end agent that can autonomously execute a DeFi strategy (e.g., yield farming with rebalancing)? - Token unlock schedule: The next major unlock is in Q3 2025. If the pivot hasn't delivered a visible product by then, expect a 50%+ correction. - Developer traction: Are new dApps being built on Zhipu's agent framework, or is it just hype?
In the meantime, remember this: Human first, hash rate second. Zhipu's founders are brilliant, but the market's hunger for a new narrative is blinding us to the reality that agents are not ready for prime time. Speed with soul, but also speed with skepticism.