Unique Research · Insight / viewpoint · January 29, 2024 · Author (CN): Tan Qiuping · Guest: Liu Yuchen, Founder & CEO, Tyrion.AI (意言科技)
If there was an AI weather vane in H2 2023, it was AI Agents. OpenAI co-founder Andrej Karpathy told an AGI hackathon in June that OpenAI’s internal attention to Agent papers had already surpassed LLM-training research — and urged builders that their Agents might sit at the frontier while even leading LLM labs were not yet ahead in the Agent layer. Why the heat? Karpathy’s answer: AGI will lean on (many) Agents; digital organizations — even “civilizations” — may emerge. Wild, and motivating.
Agents are the tool that pushes human–AI collaboration toward a higher form. Historically, enterprise-software startups needed ~18 months of fundraising while enterprise decision cycles ran a similar length; SaaS shortened that to ~13–15 months (PLG ~9–15). Today many AI-native companies can monetize from day one — and deliver value customers can see immediately. Tyrion.AI is one case: founded only ~half a year earlier, it already had shipping product and Fortune 500 manufacturing, pharma, and CPG logos, with millions of RMB in signed commercial deals.
Agents as the enterprise intelligence shift
Liu spent years in the U.S. as industry pre-sales lead and PM at a data-tech firm, running large North America finance and CPG data-platform projects (including a Forrester annual case). Back in China he led Tencent Cloud industry AI SaaS products and was CDO for L’Oréal PPD China. After ChatGPT, he saw LLMs replace labor efficiently in BA–user dialogue QC, generative BI, and analytics — at better cost/effect than classic AI — and realized one model stack could serve many business scenes. He founded Tyrion.AI to help enterprises build Agents that deeply understand human intent and jump productivity.
Versus classic bots, Liu argues LLM Agents win on (1) complex multi-step reasoning and execution — automating thinking flows, not only single tasks or fixed software logic; and (2) natural collaboration with humans, not only UI-bound interaction. Agents will seep into every “seam” of the enterprise like sand — remaking end-to-end processes until they communicate and think more like employees, lifting efficiency and agility when uncertainty is high.
Tyrion — LLM middle-layer Agent framework
Tyrion is Tyrion.AI’s LLM “middle layer” Agent framework and build platform: encode expert knowledge, embed into workflows, collaborate with employees on complex work. No-code cuts create/use friction and can shorten development time by >90%. It supports major models, light fine-tunes for small scenes, and on-prem / edge-compute packages for security-sensitive enterprises.
Five claimed edges: (1) depth — process control, complex workflow orchestration, model routing, multi-Agent collaboration; (2) usability — abstracted modules + NL→Agent workflow; (3) modularity — small models, heavy RAG, OCR/RPA hooks packaged; (4) enterprise security — privacy + edge hardware packs; (5) vertical data — domain corpora for industry Agents. The platform ships industry SOPs, vertical base data, encapsulated atomic skills (e.g. codegen), and rich configuration for large-enterprise landing.
Human–Agent collaboration at enterprise grade
Core capability splits: packaged modules for common apps, plus advanced knobs and scripts for power users. Underneath sits an internal scheduler nicknamed the “Hand of the King” — routing internal/external tools and smaller models inside one Agent flow for stability and efficiency. Platforms beat one-off packaged apps because firms can reuse and retune SOPs with minimal engineering as needs change; business owners can spin products without the old R&D/test loop.
Real enterprise work is multi-person. Replacing some roles with Agents means human↔multi-Agent and multi-human↔multi-Agent choreography: when to insert humans, when Agents, who owns which step. Tyrion’s patented multi-Agent / multi-user collaboration framework treats the flow as a human+Agent workflow — SOP, task split, feedback, eval, output — raising Agent “EQ” to read intent and smooth co-work. Customer landings already show strong fusion; scale should push deeper process innovation.
Digital employees, not just tools
Agents only fully land when treated as digital employees with process awareness. One Fortune 500 customer built an R&D assistant on Tyrion covering project management, research gathering, compare/reason, and report writing — work that once needed a junior researcher, now senior staff + Agents. Liu also notes Chinese firms are standardizing management: datafication and automation are mandatory, and Agents will be central. Tyrion’s vision: supply digital employees that handle complexity, fit idiosyncratic process, and accumulate as digital assets.
Employees gain time for creative/planning work; capable people expand into new businesses. Management focuses less on low-efficiency people ops and more on the business. But Agent exploration is early — the bottleneck is demand discovery: firms must be willing to remake processes and accept AI inside them. Vendors must both ship tech and live inside customer workflows, then productize common SOPs and general digital-employee skills.
AI-native software, not bolt-ons
Agents are software robots. LLMs + Agents reopen enterprise-software imagination so AI-native companies can carve new ground. Liu notes Salesforce, Adobe, and SAP shipped many AI features in 2023 — mostly point upgrades on existing products. Real leverage means rethinking software form AI-natively: which scattered, formerly human-only flows can machines own now?
Liu’s edge bets: years of buyer + consulting experience to pick the right process insertion points; cloud/AI team depth; energy, pharma, CPG, and media SOPs already in inventory. Customers who already tried AI tend to affirm Tyrion’s value fastest. As Agents evolve toward full-process automation, the next software shape is forming — and Liu believes Agents will birth the next enterprise software giants.
1 CN image; attach on ingest.
— END —