The first line we shipped on our website back in 2023 said: “Kiss your apps goodbye—say hello to agents.” At the time, hardly anyone understood what we meant. Two years later, the world caught up: the most valuable software isn’t another destination you click into; it’s a digital teammate that meets you where you already work, does real work on your behalf, and learns as your business learns.
This article is my working playbook—what I’ve learned across physics and enterprise software, SAP and Salesforce, and now Tribble—about where AI is going, how I think about competition, and how we’re running the company for the decade ahead.
Why now: the cost of intelligence is collapsing
When the cost of intelligence drops, business models shift. The latest Stanford AI Index 2025 quantified something we’ve felt in our own stack: inference costs for GPT‑3.5–class performance fell ~280× between Nov 2022 and Oct 2024, while hardware costs declined ~30% annually and energy efficiency improved ~40% a year. Translation: more capability per dollar, and a lot more use cases pencil out.
Adoption has followed. 78% of organizations now use AI in at least one business function, up from 55% a year prior, according to McKinsey’s 2025 survey. The companies actually seeing bottom-line impact are redesigning workflows and putting senior leaders over AI governance, not just sprinkling prompts on top of old processes.
Usage is also shifting inside the tools people live in. Slack’s workforce data shows daily AI use up 233% in six months; 60% of workers say they’re using AI, and 40% have already worked with an agent—with nearly a quarter offloading tasks to agents that complete work for them. That’s the new front door.
The interface should be conversational; the product must be the agent that acts, cites sources, requests approval, and learns from feedback. Anything less is just autocomplete glued to an API.
The interface is conversational; the product is an agent
At Tribble we build for a simple reality: work happens in Slack and Teams. Tribble shows up there—not as another dashboard, but as a teammate that reads from and writes to your systems with hard guardrails. If you’ve ever typed “YES” in all caps to confirm a destructive action, you know the pattern; we embrace similar friction so every write has provenance. When a model tried to “confirm itself,” our human-in-the-loop checks still blocked it—by design.
This isn’t a nicer search bar. It’s automation you can see and trust—what I sometimes call digital BPO. Customers hand the agent units of work; the agent brings back the completed asset and receipts. Early on we staff the edge cases with humans to accelerate learning, but the mission is to automate the play, end-to-end.
From RFPs to revenue: start narrow, go deep
We intentionally started with RFP automation. If an agent can correctly answer your RFP—from requirements to deal strategy—it has earned the right to help elsewhere in your revenue engine. By mid‑2024 the RFP market commoditized, signaling it was time to expand from a single beachhead into end-to-end GTM workflows.
Today Tribble can generate a first-pass industry deck, pull supporting references directly from Highspot/Gong/Salesforce, and route a thumbs-down back to the right owner when something’s off. When we’re confident, the agent updates Salesforce—but only after the human explicitly confirms the change inside the channel. That combination of speed and “show your work” is how you build trust.
An agent should observe the systems you care about, decide based on policy and confidence, act with citations, and learn through human feedback loops. Without all four steps it’s not an agent—it’s a macro.
Organizational memory is the moat
I don’t believe generic prompting is the future. The durable asset is the Tribble Brain: a company-specific knowledge graph that fuses what was said (calls, threads), what was shipped (decks, datasheets), and what actually happened (CRM). When someone downvotes a response, we log the source, fix it, and the improvement propagates everywhere.
You should expect agents to deliver cited answers, approvals in Slack, and automatic write-backs to the systems of record. Anything manual in that loop becomes the bottleneck.
How we run Tribble
Inside Tribble, agents are accountable for shipping work. Every critical motion—security questionnaires, partner enablement, exec briefings—runs through an agent that’s auditable, cited, and human-supervised. Leaders stay focused on strategy instead of chasing status updates.
Over the next 60 days you can turn on the same model: wire in your knowledge sources, connect Slack, set the escalation policies, and let Tribble run the first program side-by-side with your team. Backlogs shrink, approvals appear automatically, and SMEs focus on high-impact work instead of copy/paste.
Looking beyond 2025, the agent era isn’t a trend. It’s a fundamental change in how work gets done—your team can ask once, and Tribble answers forever.