Vijay Raina is a seasoned veteran in the SaaS ecosystem, recognized for his deep understanding of how enterprise architecture intersects with business ethics. As a specialist in software design and architecture, he has spent years navigating the complexities of automation and its long-term ripple effects on the global workforce. Today, we sit down with him to explore the launch of IMCeleste, an Austin-based AI customer service venture that is flipping the script on corporate responsibility. Our conversation dives into the “Dividend Standard,” a bold financial commitment to displaced workers, while examining the deterministic safety protocols that allow their AI agent to earn autonomy through a rigorous probationary system. We also explore the economic theory that automation without wealth distribution creates a structural “math” problem that could eventually leave companies with no customers left to serve.
Traditional business models frequently prioritize maximizing returns for shareholders, often at the expense of the workers whose roles are being automated. How does the philosophy behind IMCeleste challenge this standard, and why is this shift happening now?
The philosophy here is built on the realization that if we want things to be different, we have to do things differently. IMCeleste is challenging the status quo by implementing the Dividend Standard, which pledges the greater of 5% of annual revenue or 70% of annual profit to the very workers its software replaces. This is a radical departure because most companies view automation as a way to “trim the fat” and funnel those savings directly into shareholder pockets. However, the lead architect in Austin correctly points out that every person who earns a paycheck also spends one, meaning that if you remove the worker, you eventually remove the customer too. This shift is happening now because the pace of AI integration is so fast that universal basic income funded by wage taxes simply cannot keep up with the structural shock.
The decision to bootstrap and avoid outside investors is quite radical for a high-growth AI startup. What are the long-term implications of being “uninvestable” by conventional metrics while committing to the Dividend Standard?
By choosing to be bootstrapped and avoiding a traditional cap table, the company protects its core mission from being diluted by investors who would prioritize quarterly returns over worker stability. The investor note actually describes the company as “close to uninvestable” because most venture capitalists would never agree to route 70% of profits away from the owners. This structure is essential because it allows the company to honor its revenue floor, ensuring that payments are made to displaced workers even while the company is young or not yet profitable. It is a long-term play that values economic sustainability over the typical “exit at all costs” mentality seen in Silicon Valley. Without outside pressure, the firm can focus on its goal of ensuring that the gains from AI don’t just pile up in the accounts of a few providers.
When integrating AI into customer-facing roles, trust is often the biggest barrier. How does the transition from “Practice” to “Autonomous” mode allow a business to safely delegate authority to an AI agent?
The system is designed so that the AI must essentially earn its way onto the front lines through a series of observable probationary stages. Every new support topic begins in “Practice” mode, where the AI generates drafts that sit alongside the actual replies sent by the human team, ensuring that nothing reaches a customer without prior approval. Once the owner is satisfied with the performance on a specific topic, they can move it to “Co-pilot” and eventually to “Autonomous” mode. This granular control means a company can have one topic, like “Shipping Status,” fully automated while keeping more complex issues in a draft-only state. If anything goes wrong, there is a single control that can return every single topic to “Practice” mode with the press of a button.
There is a lot of talk about the “hallucination” risks of generative AI. How does Celeste use deterministic software controls to ensure that an AI doesn’t overstep its bounds when handling sensitive tasks like refunds?
The brilliance of the architecture lies in combining generative AI with deterministic, conventional code that acts as a hard boundary. An owner can write specific limits in plain language, such as “never refund an order older than thirty days” or “ask me first for refunds over two hundred dollars,” and the software enforces these strictly. These are not just “suggestions” to the AI; they are hard-coded constraints that prevent the agent from overriding human policy. If the AI encounters a risky action that hasn’t been classified, it automatically holds that case for a human to review. This ensures that while the AI can be conversational, it is never allowed to be creative with the company’s bank account or policy rules.
Beyond just chatting, this AI is designed to resolve issues through API and MCP systems. In a practical enterprise setting, how does the transparency of keeping “receipts” and showing sources change the way support teams interact with their tools?
Instead of being a “black box” that gives answers without explanation, the system provides a full audit trail for every single decision it makes. When it looks up an order or checks a customer’s qualification for a service, it keeps receipts that include the exact sources used and the confidence level of the AI at the time of the reply. This transparency allows human supervisors to see exactly why an action was taken, much like they would review the work of a new human hire. Because it is trained on the team’s past replies and provided documents, the AI becomes a reflection of the company’s existing expertise rather than a generic bot. This level of detail makes it much easier for a support lead to trust the system to handle live chat or email through Gmail and Microsoft 365.
What is your forecast for the Dividend Standard?
I believe we are at a crossroads where the “math” of the current economy will eventually force other companies to look at models like the Dividend Standard. While IMCeleste is currently an outlier by soliciting industry feedback on such a high profit-sharing percentage, the structural problems of AI wealth concentration will soon become too large to ignore. If the industry doesn’t find a way to let profits flow back to the people being replaced, the very customer base that fuels these SaaS companies will erode. My forecast is that within the next decade, we will see a “Second Wave” of AI startups that prioritize this type of economic circularity not just out of kindness, but as a necessary strategy for their own survival. Those who ignore the fact that displaced workers are also lost consumers will find themselves with the most efficient software in the world and nobody left to buy what it helps produce.
