"The AIs are talking to each other. Where's the strategy?" A conversation with Tribble's Sunil Rao

Sunil Rao spent eight years at Salesforce, much of it as a sales engineer responding to the kind of bids he now builds software to win. In April 2023 he founded Tribble, an AI-native proposal platform built around what he calls a deal brain. I sat down with him for Episode 22 of The Stargazy Brief to work through what has changed in the AI-in-proposals conversation over the past year, why he thinks speed stopped being the point, and how regulated teams should think about getting started. What follows is an edited version of that conversation.

Sunil’s answer is that the strategy lives in deal intelligence, grounding AI in the history of won and lost deals rather than a content library, so the question stops being how fast you respond and becomes how you win.

You’ve described the last couple of years as the “slop era.” What did you mean?

The “slop era” is Sunil Rao’s term for the first wave of AI proposal tools, roughly 2023 to early 2026, that generated plausible but generic responses from static content libraries without grounding them in deal context.

“In the initial phases, when LLMs were being used to generate responses, I call this the slop era. We were just generating responses. This was the first set of products that came out at the beginning. The most important thing was that they were evidence-backed, because when folks were getting these answers, it’s like asking a teammate to do it. Where’d you get that from? How did you come up with that answer? So it became really important to cite sources and give your reason for why you answered that way.

“But we’ve moved away from that now to: how can we be more strategic in the response? This is what really good bid teams actually do, the folks who are tenured in their profession. The evolution from that slop era, and the necessity for evidence, into really becoming a thought partner when you’re proposing a response, that’s something that’s played out.”

I hear the procurement side of this. Buyers tell me they can spot an AI-written proposal instantly, and it’s costing vendors deals.

“It’s happening everywhere at the same time. If you go on LinkedIn now, all of a sudden emojis are everywhere. I’ve had someone on my LinkedIn for ten years who’s never used an emoji the entire time, and now there’s a frequency of posts, and the responses are all ‘how thoughtful.’ It’s clearly AI talking to AI.

“I see the same thing on the proposal side. One of our customers was in a meeting with one of the biggest pharma companies in the world. They’d issued a massive RFP, and they were joking that their teams had used AI to generate the RFP they sent out to vendors. And the vendor was responding saying, yeah, we’re using technology to respond to it. The AIs are talking to each other. So where does the human fit in? Where’s the strategy?”

When you talk to revenue leaders now, how has the conversation changed?

"It's gone from 'I want to reduce my cost in operating this function by giving the team more leverage to respond more quickly' into 'are we bidding on the right things? Are we showing up the right way when we respond?'

"The reason we have these teams is because they're building a strategy. They're not just filling out answers in an Excel. They're doing the thought work behind how we win, behind 'this is an account that looks like the other twenty we've won, how do we position against the vendors we know are in play that aren't explicitly mentioned to us.' That's the surface area you need to cover.

"There's a class of customer whose perception is 'we just want to respond faster.' You can use an off-the-shelf AI for that, and you're competing against labs that are getting really good at it. But is that really what you want to do? You probably want to think about how you win. That's our process."

That's the deal intelligence idea. Walk me through what's actually different about it.

In short: deal intelligence grounds the AI in the full history of won and lost deals, so it reasons over what has actually won before, rather than only generating a draft from a content library.

"Take a step back. What are we trying to achieve? We're trying to win. This is one deal among many a company is pursuing, and there's a dataset of all the deals they've pursued in the past. There's behaviour that leads to a good outcome and behaviour that leads to a bad one. Now you've got a much vaster dataset than the document in front of you.

"Do you have a Mac, Chris? Time Machine. You can go back and look at a previous point in time. What if you could do that with deals? You could say, I want to go back to the deal I won for this health tech company a year ago, and I want to know exactly what the trajectory looked like, what things happened along the way that made it come out positive.

"If you combine that with the evidence you have, all the facts and knowledge relevant at the time, you start recreating how deals played out in the past. So when the AI generates a response to a specific question, the question isn't only 'document number 75 says X.' It's: did the fact that document say X help me win seven deals before? Should I say it that way again, or do something different because this looks like a deal where a particular competitor is in play? There's a level of intelligence you can infuse that's more conducive to winning than just generation."

I get pushback from bid leaders who ask why they can't just use Claude Skills for this. What's your honest answer?

"There's a bifurcation at the very beginning. If you're focused on just responding faster, go the Claude Skills route. It's super cheap, you can probably get it to generate responses, and you submit and move on. No one should do this just to be clear, but it's an option.

"When you start asking how do I win, it unfolds. What data sources do I have access to? Let's connect to your CRM and look at everything that tells us the shape of a winning deal. The CRM isn't the end-all, not everyone updates it, so what else exists? Call transcripts. Gong, Clari, Slack, Microsoft Teams. You treat that conversational data, which is noisy and time-based, differently from the Salesforce data, and you merge them. Then you've got your content, your golden RFPs, your won and lost proposals, all with signals attached.

"We call this the brain. We compile all that data and store it as a view of what happened in the past, how, and why, so the agent can ask those questions when it constructs a response. At the end of the day, why are we responding this way, and what about it makes a winning response for this customer? If you can answer that and you have the data to back it, that's what we build."

What sits inside a deal brain?

In Sunil's framing, the deal brain merges three kinds of data that most teams keep separate: transactional data from the CRM (Salesforce, HubSpot), conversational data from call and messaging tools (Gong, Clari, Microsoft Teams, Slack), and content history (golden RFPs, won and lost proposals tagged with outcomes). Connected together, they let the system reason over what has won, not only what has been said before.

A lot of the regulated teams I speak to still run everything in spreadsheets. How should they move toward AI?

"The UX becomes really important, because you have to build trust with the user. I treat this technology like an employee. When you hire someone new, you trust them to a degree, then you check their work and help them learn, and eventually you reach a point where you can give them work without watching every single thing.

"Especially in regulated industries, you have to think critically about the interface and how you surface the why and the how every time, because they're doing that work themselves anyway, manually checking source docs. You want to graft onto their existing workflow as much as possible. That gives them the confidence, or at least the explainability, on why something is said the way it is, so they'll stand behind the answer."

Most of those teams report to a CISO who says no AI. How do you get past that?

Sunil's short answer: reframe AI from an efficiency gain to a revenue gain on high-value bids, so the commercial leader, not just the security team, carries the case.

"When we started in 2023, it was difficult even in high tech to get people comfortable sending data to an LLM. CISOs were the first to ask, is this on your own infrastructure? Are you sending our data to API providers? In high tech that's more or less gone now, because the large labs have been fighting that battle. Anthropic has been on a tear. Every vertical I talk to is using Claude in some capacity. They're doing the work for us in some sense.

"In regulated industries there's more friction. In banking, one customer has an extremely allergic reaction to hosting anything on a SaaS provider, so you think about deployment models where you sit inside their infrastructure. It becomes less about the AI you're using and more about who controls it and what data leaves their VPC. Edge models becoming a play is going to be a thing as these CISOs calibrate.

"Here's the operating model that works: if the trade-off is a large amount of efficiency, it's a harder sell. But if the trade-off is a business case to improve revenue substantially, it becomes a louder voice from the commercial leader to the security team. That's why it comes back to winning more, as opposed to generating responses faster."

What does the audit trail requirement actually look like in something like banking?

In banking, regulators expect an exportable record of every decision the AI made on a response. Tribble logs each model decision in the database so it can be produced on demand.

"Anywhere you're doing anything regulatory, you're making sure you comply. We've had to think deeply about how we audit the decisions being made and the information presented to the user, logged so it's available to export. In banking specifically, the regulator will want you to be able to export essentially an audit trail of every decision that's been made. For us, there's a flag in the database, we keep this stuff, and then it becomes a question of the interface and how you surface it depending on the customer."

What best practices have stopped being best practices?

"Best practice is just another way of saying how you win. What's changed comes back to that story about AI talking to AI. Anytime someone on my team uses ChatGPT or Claude and I see the dash in the text, I have an allergic reaction. For the life of me I don't know where that key is on the keyboard. No one does. You see it and you think, this is generated.

"You use the tool, that's fine. But the customer needs to feel you've done thought work. The value of a typo, as opposed to causing a negative ROI on your response, is actually causing somewhat of a positive ROI. Because I think, that was written, a human sat there and made that typo in a series of thoughts they had, as opposed to outsourcing all of it to the AI. So best practice now is to create the surface area for people to be involved in the process, rather than handing it off completely. Someone has to be accountable for what's being shipped, because the AI can't be."

Tribble splits its product into "respond" and "engage." What's behind that?

"I used to be a sales engineer, part of the account team, going in and doing demos and discovery, attached to the deal after a certain level of qualification. Someone on the proposal team is also an extended member of the sales team. The more exposure you get to the right stage of the deal, post-qualification, the more consultative and strategic you can be in the response.

"Respond is the actual proposal generation and review. Engage is all the touchpoints before and after the proposal, where you're talking strategy with the customer and internal teams. If we remove load on the respond side with the brain and content management, the person is freed to engage more across the deal cycle. That does two things. It lets them be more strategic, and it's data collection for the pursuit, which actually informs the response and helps you win."

If I'm a bid leader who wants in on this, what should I do, even if no one's invited me to the AI conversation?

"Anything you can do to tie it back to revenue, as opposed to efficiency gains, gets to the ear of every executive up to the CEO. If a large percent of your business comes through bids and proposals, it serves the board to think about it that way. So get yourself invited is rule one."

To which my answer was a very cheeky,"... or don't get invited, just go." A lot of bid people are introverted and don't want to be in a room full of senior titles giving their opinion. But those people don't know how to win RFPs, and you do. Be in the room so they understand what your role brings, and what AI can do inside it.

For someone who hasn't really used these tools yet, what's the easiest first step?

"If you haven't already, go get yourself a ChatGPT account or a Claude account. I know it sounds basic, but these things are evolving fast. If you tried it six months ago and thought it was cute but it didn't get you what you wanted, it's changed a lot. Throw something at it, throw files at it, do research with it. You'll immediately get an intuition for how far the technology has come, and hopefully the conviction to advocate for bringing it into your organisation. It's hard to have conviction in something you haven't actually done."

Six months to a year out, what does the industry look like?

"Regulation takes a long time to catch up, so in highly regulated verticals it'll be status quo plus plus, some automation brought into the process. In high tech the appetite to experiment is far greater, so new models get explored sooner.

"Here's the question it begs. Some companies treat RFPs as a funnel of forty-five, 'I'll respond to whatever I can and see what comes back.' That's fine, some operate that way, but it's not how you win. And if you're doing it that way, do you want to staff it with people? If you've got highly paid sales engineers and you bog them down responding to something you're not best positioned to win, is that the best use of someone whose main quest is closing the deal, but whose side quest is filling out Excels? Probably not. So it begs the question of whether you'd outsource it altogether. Companies already do this for security questionnaires, with BPOs. For vendors like us, maybe the agent is just your digital BPO that answers these things for you."

You can find Sunil and Tribble at tribble.ai, and the full conversation in Episode 22 of The stargazy Brief. Tribble's profile, including how stargazy classifies it in the AI-native category, sits on the Tribble vendor page, and the wider category map is in the 2026 Proposal & Bid Software Report.

✹

FAQ

What is the "slop era" of AI proposals?

Sunil Rao's term for the first wave of AI proposal tools, roughly 2023 to early 2026, that generated responses from static content libraries without grounding them in deal context. The output read as plausible but generic, and procurement teams have started penalising vendors whose proposals look generated rather than written.

What is deal intelligence, and how is it different from AI response generation?

Response generation produces a draft faster from a content library. Deal intelligence grounds the AI in a wider dataset, including CRM records, conversational data, and the history of won and lost deals, so it can reason over what has won before rather than only what has been said before. Tribble calls the combined dataset a deal brain.

What data sources sit inside a proposal AI brain?

In Sunil's framing: CRM data (Salesforce, HubSpot), conversational data (Gong, Clari, Microsoft Teams, Slack), and content history (golden RFPs, won and lost proposals tagged with outcomes). Merged together, they let the system reason about why a given answer won in the past.

Why do CISOs in regulated industries block AI in proposals, and what changes their mind?

The concern is usually data leaving the corporate VPC, especially in banking and healthcare. Sunil's argument is that the framing matters more than the security detail: when AI is pitched as an efficiency gain, the CISO's risk weighting wins, but when it is pitched as a revenue gain on bids that drive a large share of the business, the commercial leader carries the case to the security team.

What audit trail do regulators expect from AI in bids?

In banking, the regulator expects an exportable trail of every decision the AI made on a response. Tribble's design logs each model decision in the database and surfaces it through the interface, so it can be produced on demand.

What is the easiest way for a proposal team to start with AI?

Open a Claude or ChatGPT account and use it. Throw real files and questions at it and see how it responds. Sunil's point is that the tools have changed fast enough that hands-on experience from even six months ago is out of date, and that conviction to advocate internally comes from having used them.

✹

Sources

• The Stargazy Brief, Episode 22. Interview with Sunil Rao, CEO of Tribble. All quotes are drawn from this source.

• Tribble and the 2026 Proposal & Bid Software Report (AI-native category) at stargazy.io/proposal-tech/tribble

• 2026 Proposal & Bid Software Report at stargazy.io/offers/2026-proposal-and-bid-software-report