Orange Collective
Akkari

Akkari

The AI execution system for customer operations — from the first sales call through success and expansion.

akkari.ioYYC Profile
Akkari co-founders Henry A. Lee, Jeffrey Byun, and Michael Moore with team at Y Combinator
The Akkari team at Y Combinator · P26

2x

YC founders

OrderAhead (W11) → Square; BLADE → $1.2B+/mo

Tier 1

Angel signal

Fidji Simo · Tony Xu · Eric Wu · Othman Laraki · Logan Green

Backed by

Fidji Simo

CEO, OpenAI Apps

Tony Xu

CEO, DoorDash

Eric Wu

CEO, Opendoor

Othman Laraki

CEO, Color

Logan Green

Co-founder, Lyft

SV Angel

Liquid 2 Ventures

Tuesday Capital

Y Combinator

Orange Collective

A handpicked angel list of operators who built — and scaled — the exact GTM problem Akkari is solving.

Thesis

Every customer interaction at every B2B company creates a long tail of operational work: a commitment made on a sales call, a bug surfaced in Slack, a security question asked for the third time, a fix that shipped but never got back to the customer, an expansion signal scattered across email and meeting notes. At an early-stage startup, this usually lives in people's heads. That works until it doesn't.[3]

Akkari captures every commitment, issue, request, and opportunity across the customer surface (Slack, calls, meetings, email, Discord, CRMs) and executes the actual work to close each loop — drafts the reply, files the ticket, schedules the meeting, opens the PR, sends the customer the update when the fix ships.[2][3]

Recording work is not the same as doing it. Akkari is the AI execution layer that turns conversations into completed customer outcomes.

  1. 01

    The wedge is execution, not transcription. A decade of GTM tooling — Gong, Chorus, Granola, Fireflies, Salesforce Einstein — produced excellent records of customer conversations. None of it does the work those conversations create. Akkari is the first product built around the assumption that an AI should not just summarize the call, it should finish the to-do list the call produced.

  2. 02

    The founders have lived this problem at every scale. At OrderAhead (W11, acquired by Square), Jeff and Henry built the product and Michael built the GTM org — 30-person sales team, 2,500 contractors, thousands of merchants. They learned the same lesson three times: customers are won or lost in the follow-through. At BLADE, Jeff and Henry scaled to $1.2B+ in monthly trading volume on the back of the same operational rigor.[5][6][7]

  3. 03

    The angel list is the validation. Fidji Simo (CEO, OpenAI Apps), Tony Xu (DoorDash), Eric Wu (Opendoor), Othman Laraki (Color), plus Liquid 2, Tuesday Capital, YC, and Orange Collective.[2] Every name on the list ran the exact customer-operations problem at scale and chose to back the team building the solution.

  4. 04

    Validation is happening in real time. Some of the most credible teams in the P26 batch are already customers, as well as later-stage companies including a unicorn app sec company.[3] The pull is showing up simultaneously across founder-led sales, customer success, and PLG-to-enterprise conversion — which is exactly what you'd expect from a horizontal customer-ops layer.

Problem

Every customer interaction creates a long tail of operational work. At an early-stage startup, all of it lives in people's heads.

A customer reports a bug in Slack. A founder promises a pricing breakdown on a call. A prospect asks the same security question three times. An engineer ships the fix, but nobody tells the customer. A champion is ready to expand, but the blockers are scattered across email, Slack, Linear, and meeting notes.[3]

At ten customers, the CEO holds it in their head and maybe a Notion doc and some internal Slack channels. But past that, half the team is spending half their week on the operational work created by the other half — and customers are slipping through the cracks anyway.

The conventional response is to layer in more tooling: a CRM for accounts, a ticketing system for support, a call-recording tool for sales, a customer success platform for retention. Each surface has its own database, its own AI summary feature, and its own narrow notion of what "the customer" is. None of them close the loop on the actual work.

The reason the best teams win is not better product or better messaging. They win because they follow through faster, more consistently, and with better judgement than everyone else. That's what Akkari is making available off-the-shelf.[3]

~10 calls/wk

When pre-sales breaks down

Per Akkari's onboarding guidance

~10 customers

When post-sales breaks down

Operational work stops fitting in your head

5+ surfaces

Where customer signal lives

Slack · email · calls · Discord · CRM · Linear

Why Now

For the first time, AI can do the work — not just describe it.

Two preconditions had to come true at once: companies needed a complete digital record of their customer work, and AI agents needed to be reliable enough to act on it.

The shift

Previous era · context capture

Recording the work

Gong, Chorus, Granola, Fireflies, Salesforce Einstein, HubSpot Breeze. Excellent at capturing what was said and producing a summary. Every commitment, blocker, and opportunity ends as a row in a database that someone still has to read and act on.

Now · agentic execution

Doing the work

Akkari. Drafts the reply, files the bug, schedules the meeting, opens the PR, sends the customer the update when the fix ships. The job is no longer to read the summary — it's to review the action.

AI wasn't meant to just store context or summarize conversations. It should execute the work those conversations create.[3] The model capability — Claude Sonnet 4.6 / Opus 4.7 (1M context), GPT-5, multi-step agentic tool use — finally lets a horizontal product reach across Slack, email, Linear, GitHub, and a CRM in a single coherent action. The substrate is finally there.

The proof is already in production — one channel over. The adjacent category, AI support agents, has demonstrated at enterprise scale that agentic execution on customer conversations works. Intercom's Fin reports an average 76% resolution rate across 8,000+ customers and charges $0.99 per resolved conversation — outcome-based pricing only exists when the outcome is reliable.[20] Sierra crossed $150M ARR in eight quarters and serves over 40% of the Fortune 50.[11][12] Gartner projects agentic AI will autonomously resolve 80% of common customer service issues by 2029.[19] Every one of those proof points is confined to the inbound ticket queue. The same reliability, pointed at the rest of the customer surface — commitments, follow-ups, onboarding, expansion — is the opening Akkari is built for.

One of the most experienced founder teams I've come across. These guys really know GTM. I'll be following closely!

Sherwood Callaway

Sherwood Callaway[3]

Founder & CEO · Sazabi (P26)

Proud to be customers! One of the fastest moving teams.

Muvaffak Onuş

Muvaffak Onuş[3]

Co-founder · Limrun (P26) — Akkari customer

Proud to be one of the early design partners. The guys are absolute beasts — extremely smart and experienced. Very excited to see their product grow.

Arseniy Shishaev

Arseniy Shishaev[3]

Co-founder · Superlog (P26) — Akkari design partner

Three preconditions converged in the same six-month window.

Complete digital record of customer work. Slack/Teams replaced email for internal coordination. Loom/Granola/Gong recorded the calls. Email APIs and webhooks turned every external touchpoint into structured data. For the first time, the entire customer surface — internal and external — is machine-readable. The data is sitting there, waiting to be acted on.

Agent reliability crossed the bar. A year ago, a multi-step agent that read a Slack thread, classified it as a bug, opened a Linear ticket with the right team and severity, and notified the customer when the PR merged — without a human in the loop — was a research demo. Today, on Claude Sonnet 4.6 / Opus 4.7 (1M context) and GPT-5, it's a production-grade workflow that the model can execute end-to-end with the right harness. Akkari's bet is that the harness is the product.

Builders feel the pain acutely now. Modern AI-native startups ship 5–10× faster than two years ago (the Sazabi thesis). The downstream effect: more shipped features → more customer commitments → more follow-ups → more operational work per founder-hour. The "outer loop" problem Sherwood Callaway named for observability has a customer-operations twin, and Akkari is the team building for it.

The best teams do not win just because they have better product or better messaging. They win because they follow through faster, more consistently, and with better judgement than everyone else. We're building Akkari to make that level of execution available to every startup.
— Akkari launch on Bookface[3]

How It Works

Detect signal. Triage against account context. Execute the work.

Step 01

Signal detection

Listens across Slack, calls, meetings, email, Discord, CRMs, and ticketing — surfaces every commitment, request, blocker, issue, risk, and next step the moment it appears.

Step 02

Prioritization & triage

Resolves each open item against full account context — ARR, fit score, stage, history, prior interactions, business impact — to decide what gets done now, queued, or escalated.

Step 03

Execution

Drafts the reply, files the bug into Linear with the right team, provisions access, schedules the meeting, opens a PR, sends the customer the update when the fix ships — and tracks every commitment to resolution.

Concrete loops Akkari closes today.

Pre-sales follow-through. Move a lead from a vague next step to a completed follow-up — the pricing breakdown promised on the call, the security questionnaire requested in email, the demo recording the prospect asked to share with their team — drafted, sent, and tracked.[2]

Bug-report to merged PR. Convert bug reports and feature requests from Slack, calls, and emails into Linear tickets with the right team and severity. When the PR merges, Akkari notifies the right customers — automatically.[2]

Onboarding unblocking. Turn onboarding blockers — missing API access, undocumented integrations, security review prerequisites — into queued action items routed to the right internal owner.

Quiet-churn early warning. Spot the silent signals — falling usage, missed check-ins, a champion gone quiet, a tonal shift in a thread — and trigger the save play before renewal becomes a fire drill.

Expansion play orchestration. Detect usage spikes and conversation signals indicating expansion readiness. Surface the open commitments, the unresolved blockers, and the buying signals into a single play the AE can run — or that Akkari can run with review.

Where signal comes from — and where action goes out

Integrations
SlackGmail / EmailCalls & meetingsDiscordLinearGitHubPostHogCRMsCustomer DBClaude Code / Cursor (MCP & CLI)REST API

Akkari is horizontal by design. Single-channel incumbents — a CRM AI, a ticketing AI, a meeting-notes AI — can only see one slice of the customer. Akkari sees the whole surface and closes loops across it.[2]

Akkari turning a customer conversation into a tracked feature request and shipped loop
Signal → triage → execution: a feature request closed end-to-end. (Click to enlarge.)[3]
Akkari's Slack agent triaging a customer message in a shared channel
The Slack agent acting in a shared customer channel. (Click to enlarge.)[3]

The Long Arc

A living model of how a company actually operates.

Customer-operations execution is the wedge. The durable bet is the operational memory layer that compounds underneath it.

The wedge is execution today. The compound asset is the model of the customer.

Step one — close the loops. Akkari proves it can act reliably on every customer signal: commitments, blockers, bug reports, expansion triggers, churn signals. The product earns trust action-by-action; users escalate it from drafts-only to fully autonomous on the actions where it's earned the right.

Step two — build the operational memory. Every action, follow-through, and outcome compounds into a structured record of how this customer relationship actually works: who decides, what they care about, what blockers recur, what signal precedes expansion, what precedes churn. The system gets sharper customer by customer, week by week.

Step three — become predictive. Once you have a structured model of how customers operate, you can run it forward. Which deals will close. Which accounts are about to churn. Which expansion plays will land. Where the next handoff is going to break. Akkari's framing: a living model of how companies actually operate — and one that becomes predictive over time.[3]

Akkari is our wedge towards building a living model of how companies actually operate — and one that becomes predictive over time.
— Akkari launch on Bookface[3]

Market

Three concentric markets. Akkari has commercial signal in all three already.

Near term — YC P26 and AI-native startups. The densest concentration of the buyer is in the current YC batch: technical founders who feel the customer-operations pain acutely at 10–50 customers and have authority to switch tools in a day.[3]

Mid term — Seed-to-Series B B2B SaaS. Companies that have outgrown the founder-in-everyone's-DMs phase and are starting to lose deals and accounts to follow-through gaps. Hundred-engineer-or-less GTM orgs where a single VP-Sales or VP-CS can decide. Akkari acts as an always-on operations layer across the entire customer team.

Long term — the AI operating system for companies. The category sitting between CRM (system of record) and meeting intelligence (system of capture) is system-of-execution. None of the incumbents own it. Salesforce, HubSpot, Gong, Outreach, Salesloft, Gainsight, Zendesk — every one of them has tried to add an "AI agent" surface. None of them ship execution across all channels because none of them sit across all the channels.

The software TAM is the floor, not the ceiling

Chart

AI for customer service: $12.06B (2024) → $47.82B (2030E) at 25.8% CAGR.[17] AI agents across all applications: $7.84B (2025) → $52.62B (2030E) at 46.3% CAGR.[18] Both measure software spend on the support/ticket slice. A system-of-execution prices against the labor line — the AE-hours, CS-hours, and founder-hours currently consumed by follow-through — which is why Sierra and Decagon command revenue multiples that look irrational against the software TAM and reasonable against the headcount they displace.

Source · MarketsandMarkets press releases (2025)

Competitive landscape

Four categories adjacent to Akkari. None of them execute across the full customer surface.

Each adjacent category has a structural limit. Akkari's horizontal, execution-first wedge is the answer to all four.

1. Meeting intelligence

Gong · Chorus · Granola · Fireflies

Best-in-class at recording, transcribing, and summarizing customer conversations. Strong category, deep enterprise footprint. The structural limit: these products end at the summary. The follow-up email, the Linear ticket, the security questionnaire, the expansion play — none of it gets done by the meeting tool. Akkari starts where Gong stops.

2. CRM AI

Salesforce Einstein · HubSpot Breeze · Attio AI

Smart layer on top of the CRM record. Excellent for sales analytics, lead scoring, and pipeline forecasting. Structural limit: the CRM is one surface, not the whole customer. Bug reports in Slack, security questions in email, expansion signals in Discord all live outside the CRM's field-of-view. Akkari sees the full surface.

3. Customer success platforms

Gainsight · ChurnZero · Vitally

Built around customer health scores, lifecycle stages, and playbook orchestration. Powerful for structured post-sales motions in mature orgs. Structural limit: they assume the data is already entered and the playbooks are already documented. The early-stage and mid-market reality is that neither is true — Akkari builds the underlying record by acting on raw conversation.

4. AI support agents and AI SDRs

Sierra · Decagon · Intercom Fin · 11x · Ada

The best-funded category in applied AI: Sierra raised $950M at ~$15.8B in May 2026 on $150M ARR and 40%+ of the Fortune 50; Decagon raised $250M at $4.5B in January 2026; Fin resolves 1M+ conversations a week at $0.99 per resolution. Structural limit: each is bound to a single workflow and a single channel — the inbound ticket queue or the outbound sequence. Akkari is the horizontal layer that sits across sales, success, support, and PLG — closing loops the vertical agents structurally can't see.[11]

Capital is flooding the vertical-agent layer — and converging on one channel

Chart

Post-money valuations at each disclosed round. Sierra: $175M at $4.5B (Oct 2024), $350M at $10B (Sept 2025), $950M at ~$15.8B (May 2026).[11][12][13] Decagon: $131M Series C at $1.5B (June 2025), $250M Series D at $4.5B (Jan 2026) — a 3× markup in six months, followed by an employee tender at the same price in March.[14][15][16] Hover for round details.

Source · TechCrunch · CNBC · Decagon announcements (2024–2026)

Read correctly, the capital flood is a map of where the giants aren't.

Roughly $1.3B went into Sierra and Decagon alone in the twelve months to May 2026 — and Forethought added a $25M Series D the same month.[11][14][21] Read one way, that's a crowded space. Read correctly, it's the strongest possible validation that enterprises pay real money for AI that acts on customer conversations.

Sierra and Decagon converged on inbound support because resolution rate is measurable and the deployment is self-contained. Their pricing models (per resolution) and multi-month enterprise rollouts anchor them to the ticket queue — every dollar of their roadmap is committed to resolving more tickets, faster, for bigger logos.

The unticketed majority of customer operations — the commitment made on a call, the champion gone quiet, the expansion signal buried in a Slack thread — has no incumbent and no per-resolution business model defending it. That is Akkari's lane, and the $15.8B comp at the end of the support-only path is what makes the horizontal version worth underwriting at seed.

Akkari's positioning

Akkari is the horizontal AI execution layer for customer operations. CRMs hold the record. Meeting tools capture the conversation. Support agents clear the queue. Akkari is the system that closes every loop those records and conversations create — across every channel, for every customer, end-to-end.

Founder deep dive

Three founders. One prior YC company together. One $1B+ exchange. One thesis they earned the hard way.

The OrderAhead chapter (YC W11, founded 2012 → acquired by Square late 2016). Jeff and Henry founded OrderAhead in YC W11 — a mobile ordering and payments platform that scaled out of Palo Alto into SF, LA, Houston, Denver, Chicago, and New York and was acquired by Square in late 2016. Raised from Marc Benioff, Eric Schmidt, Jerry Yang, Adam D'Angelo, SV Angel, August Capital, Matrix, Menlo, and Initialized.[9] Michael joined as one of the first 10 employees in 2012 as an Account Executive — and rode the whole arc up to VP Sales & Operations, then stayed through the Square acquisition as Sales & Success Lead managing the merchant integration. At peak he was managing a 30-person GTM team and 2,500 contractors.[4][5][6][7]

The lesson from OrderAhead — said in their own words. "At both companies, we learned that customers are won or lost in the follow-through that begins with the first conversation. The best teams do not win just because they have better product or better messaging. They win because they follow through faster, more consistently, and with better judgement than everyone else."[3] This is not a thesis they read about. It's the lesson they extracted from building one of the YC alumni cohort's most operationally complex GTM organizations.

The BLADE chapter (2018–2023). Jeff and Henry started BLADE, a crypto derivatives exchange backed by Coinbase, SV Angel, Slow Ventures, Justin Kan, and Adam D'Angelo. They scaled to $1.2B+ in monthly transaction volume — an environment where operational rigor and customer trust are literally the product. Jeff started his career as a credit derivatives trader at Morgan Stanley; the operational discipline of finance is in his bones.[5][6]

Why this is the team for this problem. Most teams attacking customer ops have either GTM pedigree or technical pedigree — rarely both, almost never with prior operating experience at the same company. Jeff and Henry have shipped two companies together (one acquired, one $1B+ in monthly transaction volume). Michael has run the customer-ops function at every scale from 10 to thousands. They're not learning the problem; they're building the product they spent two companies wishing existed.

The angel list is a recommendation letter. Fidji Simo (CEO, OpenAI Apps), Tony Xu (DoorDash CEO), Eric Wu (Opendoor CEO), Othman Laraki (Color CEO), plus Liquid 2 and Tuesday Capital.[2] Every name on the list ran the exact problem at massive scale: DoorDash on marketplace ops, Opendoor on transactional follow-through, Color on healthcare account management, OpenAI Apps on the next-generation execution layer. They're not investing in a deck; they're investing in the team they wish they'd had.

How they run the playbook. Tight design-partner cycle inside P26.[3] Founders directly in the customer Slacks. Public, opinionated launch with a clean wedge-and-vision split. The same "follow through faster, more consistently" rigor they're selling, applied to their own GTM. The compounding starts at the founder loop.

Founders

Jeffrey Byun

Jeffrey Byun

Repeat FounderExited

Co-founder & CEO

2x YC founder (W11, P26). Previously co-founded OrderAhead (W11) — a mobile ordering platform that scaled to millions of users and thousands of merchants before being acquired by Square. Then founded BLADE Holdings, a crypto derivatives exchange backed by Coinbase, SV Angel, Slow Ventures, Justin Kan, and Adam D'Angelo, which reached $1.2B+ in monthly trading volume. Started his career trading credit derivatives at Morgan Stanley. MIT and University of Pennsylvania.

Henry A. Lee

Henry A. Lee

Repeat FounderExited

Co-founder

Co-founded OrderAhead (W11) as CTO with Jeff — acquired by Square. Then CTO/co-founder at BLADE. Earlier engineer at Increo Solutions (acquired by Box). Stanford CS. Deep technical operator across two prior YC companies with Jeff.

Michael Moore

Michael Moore

Repeat FounderExited

Co-founder

Joined OrderAhead in 2012 as one of the first 10 employees and grew with it through acquisition — Account Executive → Manager (Merchant Growth) → Head of Sales & Success → VP Sales & Operations → Sales & Success Lead at Square post-acquisition. Managed a 30-person GTM team and 2,500 contractors at peak. Later COO at PlayVision and VP of Operations at Sittercity. UNC Chapel Hill. Has lived the customer-operations problem from the inside at every scale.

Risks & mitigations

Risk

Execution accuracy — when an AI sends the wrong follow-up to a customer, the cost is a damaged relationship, not just a wasted token.

Mitigation

The team treats every action as either review-required or fully autonomous on a per-customer, per-action basis. Drafts vs. sends, files vs. resolves, alerts vs. acts. The product is built around the assumption that trust accrues action-by-action and is calibrated by the user — not granted up front.

Risk

Slack, Linear, Notion, Salesforce, HubSpot, or another incumbent ships a native version of this and absorbs the market.

Mitigation

The wedge is horizontal-by-design: Akkari sits across Slack, calls, email, Discord, Linear, GitHub, PostHog, and CRMs simultaneously. Single-system incumbents can't credibly close the loop across the whole customer surface because each is locked to its own channel. The defensible product is the cross-channel memory and the act-on-the-customer execution layer — neither of which sits naturally inside a CRM or a ticketing tool.

Risk

ICP discipline — "customer operations" is broad. The pull is showing up in sales, success, support, and PLG conversion. Spreading thin is the obvious failure mode.

Mitigation

The team is running a disciplined design-partner motion across AI-native startups and a unicorn app sec company with weekly check-ins. They're seeing pull across sales, success, and PLG and explicitly choosing to keep the surface horizontal during the wedge phase — because the operational pain is the same regardless of which team owns it. Tightening the ICP becomes a Series A problem; right now the discovery is the moat.

Risk

Customer-data privacy and access risk — Akkari needs to read calls, emails, Slack, Discord, and CRMs to work. Enterprise procurement on this scope of access is slow.

Mitigation

Initial wedge is AI-native startups where a single technical founder is the buyer and the procurement cycle is a day, not a quarter. Sazabi (also P26) is following the same wedge — sell to your batch first, then expand outward.

What we're watching

  • ICP focus — does the team lean further into one of (founder-led sales / customer success / PLG-to-enterprise conversion), or do they sustain the horizontal customer-ops wedge?
  • Where execution accuracy lands in production — false positives on commitments, sent-without-review actions, and how quickly the trust loop with customers tightens.
  • Whether the founders use Akkari's own operational memory layer to build a self-improving, predictive model of how their customers actually operate — the long-arc bet.

References

  1. [1]Akkari — YC Profile
  2. [2]Akkari — Company Website
  3. [3]Akkari — Launch on Bookface (YC internal, P26)
  4. [4]OrderAhead (W11) — YC Profile (acquired by Square)
  5. [5]Jeffrey Byun — LinkedIn
  6. [6]Henry A. Lee — LinkedIn
  7. [7]Michael Moore — LinkedIn
  8. [8]Jeffrey Byun on X (@jeffbyun)
  9. [9]OrderAhead — Wikipedia (founded 2012, acquired by Square late 2016, $10.5M total raised)
  10. [10]Akkari — Launch YC (public launch post)
  11. [11]Sierra raises $950M as the race to own enterprise AI gets serious — TechCrunch (May 4, 2026)
  12. [12]Bret Taylor's Sierra raises nearly $1B in latest AI capital push — CNBC (May 4, 2026)
  13. [13]Bret Taylor's Sierra raises $350M at a $10B valuation — TechCrunch (Sept 4, 2025)
  14. [14]Decagon's $250M Series D at $4.5B — Decagon (Jan 28, 2026)
  15. [15]Decagon raises $131M Series C at $1.5B valuation — Decagon (June 23, 2025)
  16. [16]Decagon completes first tender offer at $4.5B valuation — TechCrunch (Mar 4, 2026)
  17. [17]AI for Customer Service Market worth $47.82B by 2030 (25.8% CAGR) — MarketsandMarkets
  18. [18]AI Agents Market worth $52.62B by 2030 (46.3% CAGR) — MarketsandMarkets
  19. [19]Gartner: agentic AI will autonomously resolve 80% of common customer service issues by 2029 (Mar 2025)
  20. [20]AI customer service agent pricing comparison (Fin: 76% avg resolution across 8,000+ customers, $0.99/resolution) — Fin.ai
  21. [21]Decagon, Sierra AI and the race to build customer support agents — Upstarts Media (May 2026)