AI-Powered Revenue is Here

In this episode of Founder Firesides, YC General Partner Diana Hu talks with Ali Akhtar (Co-founder & CEO) and Armen Forget (Co-founder & CTO) of Letter AI, who just announced a $40M Series B. Letter AI is an AI-native sales enablement platform that helps revenue teams ramp faster, generate personalized buyer content during live deals, and practice high-stakes conversations before they happen. After pivoting during YC, the company landed enterprise customers like Lenovo in the batch and has since expanded rapidly. They discuss what they learned from the pivot, how they closed major customers early, and why AI is reshaping the future of sales.

Akhtar & Forget / YC — Letter AI for AI-powered revenue

Diana Hu (YC General Partner) interviews Ali Akhtar (CEO) and Armen Forget (CTO) of Letter AI on YC Root Access — Founder Firesides episode, published 25 February 2026, ~10 minutes. Frames the $40M Series B announcement and the company’s pivot-during-batch story; Letter AI is an AI-native sales-enablement platform for revenue teams at enterprises like Lenovo, Adobe, Novo Nordisk, Plaid, and Kong.

TL;DR

• Pivot-during-batch with original-name shed: Letter AI was originally Tractatus, building developer tools for generative AI. The space saturated quickly; the SaaS-to-developers motion was non-sticky; the team pivoted during YC and “changed the name as well to make it a little bit more memorable.” Lenovo closed in the batch — “which is very rare” (Hu).
• The legacy-enablement-tool-adoption-gap insight: Akhtar’s prior-role observation at Samsara — sellers couldn’t find content in the “really expensive” legacy enablement tool (“I can’t find anything in there. I barely log in”); he, as Director of Engineering for ML, was getting pinged weekly by sellers to come give product talks. The wedge: replace the failed-content-curation layer with AI-native personalised content generation that scales without throwing humans at the curation problem.
• MCP-server + agent-to-agent protocol as B2B product surface: Letter AI’s tech-architecture surface includes “our own Letter AI MCP servers” and “agent-to-agent protocol that allows our customers — either from their web applications, their agents to talk to ours for distributed thinking”. Salespeople in Cursor can pull content via the Letter MCP server during their research workflow. First wiki source articulating MCP-server-as-vendor-product-surface for a non-developer-tools B2B vertical.
• Letter Compass = personalised-to-book-of-business: a new product layer that “takes the broader enablement content… pitch decks, training, certifications, role plays… and automatically personalises it to the book of business that a particular seller or a customer success manager owns.” Tied to conversational-intelligence and CRM data. Akhtar’s tagline: “never sell alone with Letter.”
• The 100% daily adoption claim and the Fortune-100-acquisition-onboarding anecdote: customer testimonial that pre-Letter onboarding hundreds of new sellers post-acquisition would have taken “at least a month and tons of folks”; with Letter, “they were able to do it over a weekend with just about two or three folks online.” Unaudited founder anecdote — directional, not measured.

What was actually ingested

The full ~10-minute transcript (auto-generated English captions, ASR-cleaned). Light cleanup: “Alli”→“Ali”, “Arman”→“Armen”, “Tractatus” standardised (host said “Trackus” once; cofounder’s Tractatus used as canonical), “letter AI” → “Letter AI” in body, “customerf facing” → “customer-facing”, “agentto agent” → “agent-to-agent”. Stage cues retained. Full raw at raw/videos/2026-02-25-akhtar-forget-yc-letter-ai-powered-revenue.md.

Substantive content

1. The pivot template — Tractatus → Letter AI

Forget narrates the pivot rationale in two sentences: “we were basically doing developer tools for generative AI… the field was getting very saturated. We were trying to sell SaaS tools to developers who wanted to write Python code, and they would very quickly do prototyping in our platform and then they would just go build it themselves. And so it wasn’t very sticky.” The pivot wasn’t framed as ideologically driven — “developer tooling is unsticky for early-2024 LLM-product cohorts” is presented as an empirical learning from talking to customers. Akhtar then drives the new-idea selection, anchored on his own prior-role pain.

2. The legacy-enablement-tool-adoption-gap insight (3:08–4:37)

The load-bearing customer-discovery moment. Akhtar, as ML Director of Engineering at Samsara and project44, observed:

1. The legacy enablement stack was “so expensive” that even he couldn’t get a license.
2. Adoption was so low (“I barely log in” — sellers) that sellers were pinging Akhtar weekly to come give product talks rather than self-serve from the tool.
3. The legacy-vendor’s get-leverage-from-tool answer was throw more humans at content curation — a service motion masquerading as product.

The AI-native wedge: tap existing knowledge sources (CRM, conversational intelligence, product docs) and generate personalised content in-context, without the manual curation layer. Akhtar: “speed is what matters in today’s day and age for high-velocity sales organisations.” The wiki’s clearest founder-vantage articulation of AI-native replacement of a vendor-category that failed via adoption-deficit, not via missing features.

3. The MCP-server + agent-to-agent product surface (8:23–9:16)

Forget: “we’ve been noticing a lot of our customers are becoming also very AI-centric, and they’re building their own internal tools or even customer-facing tools that are using AI — like agents — and they’re now using MCP servers and things like that. So we’ve been building our own Letter AI MCP servers. We have agent-to-agent protocol that allows our customers — either from their web applications, their agents to talk to ours for distributed thinking, for example, or their salespeople now are sometimes in Cursor and they’re doing their own research, and now they can use a Letter AI MCP server to get the content or get answers to the questions they need.”

Three distinct surfaces are bundled in those two sentences:

1. Letter AI MCP server — exposed to customers’ agents and to sellers using coding tools (Cursor mentioned by name) as a content-retrieval-via-tool-call path.
2. Agent-to-agent protocol — Letter AI’s agents talk to customers’ agents for “distributed thinking” (framing is loose; implementation not specified on stage).
3. In-line content delivery during deal cycles — Letter Compass surfaces content during live deals, tied to the seller’s specific opportunity and the buyer’s CRM context.

This is the wiki’s first founder-vantage worked example of an AI-native B2B vertical-SaaS vendor treating MCP server and agent-to-agent protocol as first-class product affordances, not as developer-platform integrations. The product surface is structured for the customer-side agent ecosystem from the outset.

4. Letter Compass — personalisation-to-book-of-business (7:20–8:23)

The newly-announced product. Takes the broader enablement assets (content, pitch decks, training, certifications, role plays) and “automatically personalises it to the book of business that a particular seller or customer success manager owns.” When a seller logs into Letter, the surface is “no longer a generic training on XYZ product. It’s a training that’s relevant to the deal that you’re pursuing today.”

Inputs named: “conversational intelligence data and CRM data.” Outputs: contextualised content + insights + follow-ups “to help you really push the deal forward.”

Akhtar reframes this as a category-redefinition move: “it forces you to rethink: what does the future of CRM look like? Is that really what customer relationship management should be — where AI is presenting insights for you to motivate and move a deal forward?” The implicit claim: in the AI-native era, CRM stops being a system-of-record and becomes a system-of-suggestion. Not load-bearing as analysis (Akhtar gestures rather than argues), but worth flagging as a named-by-the-vendor categorical pivot in the sales-enablement-to-CRM stack.

5. The Lenovo-closed-in-batch anchor (2:42–4:52)

“You were actually able to close Lenovo as a customer in the batch, which is very rare” (Hu). The mechanism Akhtar narrates: a former colleague at one of Akhtar’s prior companies “was kind enough to see what we were trying to do, and it lit a light bulb in his mind, and he connected us to the right folks within sales.” Once on-stage with the Lenovo team, “a few steps to make sure we were enterprise-ready early on in the journey” unlocked the deal. Has since “grown 10x over the course of the past two years.”

The same operational pattern Dinakaran narrates with Cleveland Clinic (Luminai): warm intro → champion identification → trust-building → enterprise-readiness work → expansion. The Letter AI version is foreshortened (told in ~30 seconds rather than ~8 minutes) but rhetorically identical.

6. The 100%-daily-adoption claim and the Fortune-100 anecdote (5:38–6:54)

Akhtar’s two adoption claims:

• “Particularly compared to the legacy tools where adoption can sometimes be less than 50%, we see customers with close to 100% adoption.”
• Customer anecdote (Fortune 100 acquirer): “They announced that acquisition internally on a Friday, and by Monday they had a whole certification in terms of ramping up these new sellers onto the acquiring company’s way of doing business. What they told us is pre-Letter, an exercise like this would have taken them at least a month and tons of folks to get done. And with Letter, they were able to do it over a weekend with just about two or three folks online.”

Both are unaudited founder-vantage anecdotes. The 100% figure is “close to 100%” not measured; the Fortune-100 anecdote is a single customer story without comparative baseline.

7. Operational scale and capital signals

• Latest round: $40M Series B announced 25 February 2026.
• Customer mix (named on stage): enterprises — Lenovo, Adobe, Novo Nordisk; growth-stage — Plaid, Kong.
• YC batch: ~2.5 years prior to fireside, i.e. roughly summer 2023 (W23 or S23 — not specified on stage).
• Original name: Tractatus.

Linked entities and concepts

Entities directly named or substantively discussed in the source:

• Y Combinator — accelerator; Letter AI’s batch (W23 or S23). Source-count bumps 8→9.
• Diana Hu — interviewing GP. Source-count bumps 2→3.
• Letter AI — Dangling first mention (the company; was Tractatus).
• Ali Akhtar — Dangling first mention (CEO / co-founder).
• Armen Forget — Dangling first mention (CTO / co-founder).
• Lenovo — Dangling first mention (closed-in-batch anchor customer).
• Cursor — substantive mention (the surface from which sellers retrieve via Letter MCP server). Existing entity if it has a page; otherwise Dangling.

Concepts touched substantively:

• enterprise-ai-adoption — adoption-deficit-of-legacy-vendor as wedge; AI-native replacement; 100% daily adoption claim. Source-count: +1.
• agent-harness — MCP-server-as-vendor-product-surface + agent-to-agent protocol; Letter AI as a node in customer-side agent ecosystems. Source-count: +1.
• agentic-engineering — vendor-product-shape implications of MCP-server-first design. Source-count: +1.

Dangling (single-source mentions, deferred): Letter AI, Ali Akhtar, Armen Forget, Lenovo. (Note: Cursor is likely a multi-source entity already in the corpus via prior ingests; treat as existing entity if so.)

Caveats

• 10-minute YC Root Access Founder Fireside with overt Series B announcement framing; financial and adoption claims are founder self-reports, unaudited.
• The “close to 100% adoption” and the “weekend with 2-3 folks online” anecdotes are unverified single-customer stories without baseline measurement.
• The 10x revenue-growth figure for the Lenovo deal is unaudited.
• The agent-to-agent protocol implementation is not specified on stage; “distributed thinking” is a vendor-marketing phrase, not an architecture description.
• Letter Compass is presented as launching; on-stage product description is forward-looking, not retrospective.
• ASR ambiguity on the original company name — the host calls it “Trackus”, cofounder calls it “Tractatus”. The wiki uses Tractatus per the cofounder’s own usage.
• Brevity caveat: at 10 minutes, this is one of the shorter sources in the YC-channel cluster; the substantive density is correspondingly lower than Luminai (33 min), Momentic (34 min), or Garg (5:34, but more conceptually packed).