It's 3 a.m. Your live-service operations team is fast asleep, but your AI Agent detects that 7-day retention has dropped 12% and begins analyzing the problem. What channel? Which app version? Which cohort? Cross-referencing internal meeting notes and external user feedback, the Agent can pinpoint the cause in under 30 minutes. It looks like a new onboarding flow released last week is causing players to get stuck at a critical level. The Agent generates a plan to optimize this flow and launches an A/B test. Two hours later, the Agent selects the winning variant and begins rolling it out to all users.
Hours later, your team wakes up and is greeted by the Agent's daily report: This problem has been solved overnight.
This isn't science-fiction, it's ThinkingAI's promise of how Agents can drive business growth with Agentic Engine, a new enterprise-level AI Agent platform. Here's a breakdown of what we see is happening in the Agentic AI space, and how our enterprise solution differs.
This is not a sci-fi scenario, but a new model of business growth driven by AI Agents. On the afternoon of April 16 (PT), ThinkingAI hosted a product launch event at the Computer History Museum in Silicon Valley, officially unveiling its enterprise-grade AI Agent platform — Agentic Engine.
1. Anyone Can Have an Agent. But How Much Does Your Agent Actually Do?
Over the past year, we have met with dozens of companies across industries and discovered that most enterprises are still only partially integrating AI by utilizing LLMs for rudimentary Q&A functions. Even teams that have begun building out Agent teams internally face numerous bottlenecks that keep them stuck on the ground level. They deploy Agents, but can't yet make them work for their business.
Here's why.
- Isolated Agents with no collaboration. An issue we found is when businesses silo their Agents within departments. The Agents don't communicate, don't collaborate, and can't share important context with each other. The Agents are in place, but the organization hasn't given them the chance to work.
- Agents are Siloed From Accessing Important Business Context. We find that Agents need more than just structured behavioral data. They need context. Community comments, support tickets, internal conversations, images, and videos. All of these are datapoints that are scattered across different systems, siloed, and invisible to Agents and therefore useless. Give Agents access to this data, however, and suddenly you change the game.
- Turning Data Into Knowledge Is Hard. Collecting data is the easy part. Agents require your raw data to be structured in order to be understood and called upon. Otherwise, your Agents are merely skimming the surface without making genuinely insightful judgments informed by your business.
- What About Security? When Agents begin acting autonomously the question of "who can see what data" and "who can take what action" become the most important ones to ask. Permissions, compliance, auditing, until these problems are solved Agents remain trapped in the "chat assistant" stage and cannot truly drive business outcomes.
2. Agentic Engine: Detect, Decide, Execute
We built Agentic Engine to answer the question: "What should a great Agent be capable of?" Leveraging our decade of experience in the data intelligence world, our self-hosted enterprise AI Agent platform focuses on three words: Detect, Decide, and Execute. This is the design philosophy of Agentic Engine.
Detect: Real-Time Alerts
24/7 signal monitoring across all your channels. Not just alerting metric anomalies on dashboards, but tracking user complaints on Discord and X, a sudden spike in negative App Store reviews, or even stalled discussions in internal meetings. The agent automatically collects these scattered signals and determines whether something is an isolated incident, or a systemic problem and delivers real-time alerts based on its conclusion. The agent doesn't wait to be asked, it proactively discovers problems.
Decide: Generates Optimal Plan
The Agent doesn't just know what happened, it knows why. When the business asks, "why did retention drop this week" the Agent knows which retention circulation formula your company uses, how to break it down by channel, version, and cohort, and knows how similar problems were resolved in the past. It functions like an expert member of a team, capable of making accurate judgments based on a full contextual understanding of your business.
Execute: One-Click Approval
The Agent creates a strategy and executes it directly. For example, the Agent can detect that ROI for a particular channel has been declining. Instead of waiting around, the Agent can automatically generate a strategy to reduce and reallocate the budget by launching an A/B test to validate its hypothesis before rolling out the winning result. No need to manually schedule anything, the Agent completes the full loop from decision to action instead of producing a report and waiting on human intervention.
Detect, Decide, Execute: a never-ending intelligence loop.
3. A Collaborative Agent Team with Unique Roles
These aren't just personal AI assistants, Agentic Engine can create an entire team of Agents who can collaborate with each other, and your employees. Agentic Engine comes with three pre-built Agents.
Analytics Agent — The 'Eyes'
Previously, an AI data analysis process began with a user submitting a request to an analyst. The analyst then writes the request in SQL, generates a report, and calls a meeting. This is a full analysis cycle measured in days. With Agentic Engine, users can ask the Agent their request directly and receive a report and action recommendation plan in minutes.
A/B Experiment Agent — The 'Judge'
A typical experimentation timeline takes 2-4 weeks and involves developing, scheduling, launching, and manually reviewing the results. The A/B Experiment Agent can recognize potential opportunities, automatically generate a hypothesis, launch the test, and monitor the experiment in real-time, delivering satisfying results without human intervention.
Engagement Agent — The 'Hands'
Automatically generates strategies from insights and delivers them with precision. Your Engagement Agent is able to detect real-time signals for at-risk users and find ways to re-engage high-value users based on behavioral signals, compressing the operations cycle from a weeks to minutes.
Create Your Own Agent
No code required. While Agentic Engine comes pre-packaged with a variety of Agents built on top of 10 years of live-service expertise, users can create personalized Agents through a simple, drag-and-drop interface. Custom Agents can collaborate with our native Agents just like any other team member.
Our Agents don't just work in isolation. Behind the Agent team is a three-layer coworking mechanism.
Strategy Layer: The Insight Agents can proactively surface anomalies and discover opportunities while the Experiment Agents can automatically validate tests. Agents don't just respond, they take initiative.
Orchestration Layer: The "brains" of the system. A unified Orchestrator handles task scheduling, state management, and context sharing. Without the Orchestrator, the Strategy and Execution layers are basically two isolated systems.
Execution Layer: The Strategy Layer runs multiple business Agents in parallel to complete specific actions, whether that's adjusting ad spend, reaching users, or responding to customers. Every result from the Execution Layer feeds back into the Strategy Layer, making it smarter and faster. This is what separates a collection of AI assistants from an Agent team that collaborates with each other, learns, and evolves.
4. Ten Years of Experience Makes Our Agents Business Experts, Not Just Generalists.
Many other Agent platforms can assemble Agents. But the real question is whether your Agents truly understand your business?
General purpose LLMs have many uses, but they don't know what formula your company uses to calculate "retention," don't know the specific definition of "new user" in your system, and don't know the dozens of ways "payment analysis" can be broken down. This institutional knowledge cannot be bridged with prompts alone.
Over the past 10 years, ThinkingAI has served over 1,500 enterprises and over 8,000 products across gaming, social, e-commerce, short-form drama, live-streaming, and other industries. We have taken this knowledge and converted it into Agentic Engine's three-layer knowledge architecture:
Layer 1: The Agent Comes With Your Business's Knowledge
Traditional data warehouses are designed for human analysts who know SQL and table structures. Agents need a knowledge base that directly understands business language. Context like how 'DAU' is calculated, whether "last week" means calendar week or operational week, if "revenue" means GMV or net receipts, all of this is knowledge your agents can ingest to inform its decisions.
Layer 2: Over 100 Pre-Built Skills
Our Agents are able to access over 100 pre-built, industry-specific skills. Skills that include User Analysis, Retention Analysis, Payment Analysis, Advertising Analysis, and Operations Analysis, these aren't generic data-query capabilities. These are skills tailored to the data industry and allow our agents to understand terms like "retention" or how to properly attribute "ad ROI."
Layer 3: The Agents Will Evolve
Every result from an executed action becomes "new knowledge" that feeds back into the system. The A/B test result, which users responded to which campaign, which anomalies are actual problems rather than normal fluctuations. Rather than resetting from scratch each time, Agents operate on an ever-growing knowledge base that becomes more accurate over time. Users can also encode their own experience into proprietary Skills which in turn makes Agentic Engine a platform capable of absorbing an enterprise's unique knowledge for continuous evolution.
5. Trustworthy, Controllable, Privately Deployable
Autonomous Agents can cause some enterprises to be nervous. There are instances where Agents act like a Black Box with no way to view how the Agent is spending, or ways to verify its output accuracy or whether the Agent is actually solving a problem or just spinning in circles. Agentic Engine avoids this by providing full-chain observation so that every action is logged and accounted for.
Sandbox Isolation: New Agents run in a sandbox without affecting the product environment.
A/B Gray Release: New and existing Agents are compared and validated and only the winners are tasked with full deployment.
Data Definition Consistency: The same questions will always yield the same answers.
Hallucination Detection: Runs throughout the full pipeline to prevent errors.
Finally, data never leaves your business. Our self-hosted private deployment is fully compliant with global privacy laws. Agentic Engine also natively supports the MCP and A2A protocols and can integrate seamlessly with any AI platform. While MiniMax is our official partner and provides the default LLM, we believe a truly enterprise-grade platform must be open.
That means the system supports Slack and other major workplace platforms, allowing users to interact with Agents anytime, anywhere.
6. The Next 10 Years
Over the past 10 years, ThinkingAI has provided data infrastructure for enterprises. Over the next 10 years, we aim to help every enterprise build its own AI Agent team. Humans will always set the goals and define the boundaries of your business. And Agents will operate autonomously within those boundaries. We believe humans will always handle strategy, creativity, and quality control. But Agents can take on providing insights, analysis, and execution. Each team member, human and AI, will contribute their strengths to the enterprise. This is our AI philosophy in the Agent era and the design philosophy behind Agentic Engine.
Effective today, Agentic Engine is officially open to customers worldwide.
© 2025 ThinkingAI · Originally published on WeChat Official Account · Translated and localized to English
