
Quick note before we dive in: we’re recruiting for Parallel’s GTM team right now (SAs, sales leaders, + Ent AEs in SF/NYC/Austin/London). If you want an intro, shoot me an email at [email protected] or reply to this email.
As some of you know, we run a GTM recruiting firm @ Crew.
The most common question we get is "how do I know the best company to choose?"
It's probably the most important question to ask. Specific to sales, we've written articles on this before (see 12 traits that make a great sales job).
But I've been really inspired by newsletters like Why You Should Join and Next Play's Spotlights, and I kept thinking: none of these are doing this specifically for go-to-market.
The reason why the distinction is important: a company doing well (press, funding, customers, etc.) and a company being a great place to join on the GTM side are two very different things.
And also, as much as Chris Orlob wants to disagree with me, the company you join is 10x more important than actually being good at sales (which you should definitely still try to do).
So we wanted to create these narrative pieces. Almost like investment memos, but easier to read.
Breaking down:
The problem the company is solving + origin story
The market the company is going after
The product the company is building to solve the problem for the market
The team that you'll be joining
Why you should actually join (on the sales team specifically)
We did Basis back in February (the "Harvey for accounting"). Then Avoca in April (the AI workforce for the trades). Then Listen Labs in June (the AI-native customer research platform). Then Blacksmith in October (the CI cloud).
Today: Parallel.
Quick background on us + Parallel: I sold Metronome to Travers (their co-founder) back in 2024, loved what they were building so much I wrote about it here, and we’ve helped build their GTM team since. Graham has been on the pod, Liam has been on twice, and they’ve sponsored the show.
So yes, I’m biased. I also know this team better than almost any we’ve covered.
Parallel’s founding theory: the main user of the web is now AI agents, not people.
Agents will use the web more than humans ever have, and every one of them needs fresh, accurate info from the outside world to do real work.
So the question for anyone building with AI becomes: do your agents need the web? [Obviously] they do; a model without it is a smart employee with no internet access.
Parallel is the infrastructure that gets it to them, built for machines instead of people clicking links.
It was founded by Parag Agrawal, the former CEO of Twitter, and the numbers are nuts:
$100M Series B at a $2B valuation led by Sequoia (April 2026), just five months after a $100M Series A at $740M co-led by Kleiner Perkins and Index. $230M raised total.
Launched publicly in August 2025. So: roughly nine months from public launch to $2B.
100,000+ developers building on the platform.
The customer wall: Harvey, Notion, Lovable, Replit, Vercel, Granola, Pfizer, LangChain, Modal, Rogo, Opendoor, Profound, and Clay, plus Fortune 50 companies, banks, and hedge funds it doesn’t name.
All of that with a team that was about 50 people in April and has 2x’ed since.
Let's get into it.
What We'll Break Down

The Problem
AI got built in layers, and the last one is still up for grabs.
Layer one was training: the labs had models read a huge share of the public web and compress it. Layer two was reasoning, which made those models really smart.
But even the most powerful model in the world is a snapshot. It was trained on the web, but it doesn’t have its own copy of the web, and it’s already two to six months out of date the day it ships. Ask it about anything newer and it has two options: go outside itself to look it up, or guess (or what the non-technical would call hallucinate).
Layer three is retrieval: getting agents fresh, checkable info from the live web at the moment they act. Liam Mulcahy (Parallel’s Head of AI Native) lays this out in The outside world is the new infrastructure, and one line sums it up: without retrieval, a model is “working from a plane with no Wi-Fi.”
The catch is that the web isn’t ready for this user. The entire web was built for one user: humans.
Agents don’t click links. They need the actual information, pulled out and formatted so it drops straight into the model.
Legacy search fails agents in three ways:
Format: links instead of answers.
Freshness: models only know what they were trained on, and an agent doing real work (underwriting, legal research, account prep) needs what's true today.
Economics: shoving whole web pages into a model burns tokens. For those less technical, tokens are the units AI models bill by, roughly chunks of words. More junk in = bigger bill.
Agrawal's framing to Reuters is the whole pitch in two sentences:
"How many jobs are there where we could turn off web access and ask you to do the same job fully? You can't deprive an M&A lawyer from not being able to use the web, so why would you deprive their agents?"
And the timing got sped up. Microsoft retired the Bing Search APIs in August 2025, pulling the default supplier out from under developers. Meanwhile, per Parallel, 79% of major news sites now block at least one AI crawler. The web is closing to machines right as machines become its biggest user.
Someone has to rebuild the plumbing - aka Parallel.
The Market
TAM is a huge predictor of your growth at a company - a mistake I've made before choosing sales gigs 🙂.
Google’s “Search & other” ad revenue was $224.5B in 2025, on 5 trillion+ searches a year. That’s what web search is worth with one type of user.
Now look at the second user. By Liam’s math, ChatGPT alone already pulls roughly 7 trillion web pages a year. That’s one product already past Google’s total, before you count Claude, Grok, Gemini, or the agents companies are building themselves.
And Parag thinks that’s the very beginning. His line on Sourcery in October: “I was the lunatic walking around 3 years ago telling people agents will use the web 1000x more than humans ever have.”
The layers above retrieval are already priced: about $2.2T for the eight leading labs, $291B for the industry AI apps, $112B for inference. The retrieval companies combined: about $29B.
The category money in the last nine months:
Tavily was acquired by Nebius for $275M (Feb 2026).
Parallel hit $2B led by Sequoia (April).
Exa raised $250M at $2.2B led by a16z three weeks later (May), and says it serves 400,000+ developers.
This is a platform race, not a feature. And it's competitive (usually a sign of a good market): Exa, Tavily, Perplexity's API, and the search built into the model labs' own products.
The obvious bear case: OpenAI, Anthropic, and Google could just do search themselves. The counter is the Stripe/Twilio playbook, where specialist infra beat the giants' in-house versions. And the tell: Google itself made Parallel a grounding option in Gemini Enterprise and Google Cloud (July 2026). When the platforms distribute you, that's a strong signal.
The Product
One-line version: one proprietary web index, and a set of APIs that let agents search, read, research, and watch the web.
For those less technical: an API is just a way for one piece of software to ask another for something, and a web index is Parallel’s own giant, constantly updated copy of the web (billions of pages, millions added daily), so it isn’t renting Google or Bing.
What you get (product overview):
Search: fast, cited web results for an agent.
Extract: turns any URL (PDFs, government sites, messy pages) into clean text.
Task: deep research. Give it an objective, get back a structured, cited report.
FindAll: "find me every X" in plain English, returned as a live list.
Monitor: watches the web and pings you when something changes.
Decisions: launched Oct 25, quick yes/no-style calls grounded in the live web, at $1 per 1,000 calls.
The enterprise wedge is Basis: every output carries citations, reasoning, and a confidence score on each field. That's why banks run KYB checks (know-your-business, aka vetting a company before doing business with it) on it. Add SOC 2 Type II, zero data retention, and HIPAA readiness (the security certifications procurement asks for before anyone signs), and regulated industries can actually buy.
Pricing is per request, not per token, starting around $1 per 1,000 searches.
October was a busy month:
Browserbase default search (Oct 14)
EU data residency (Oct 22), which opens European enterprise deals
Parallel for Finance (Oct 29)
A Slack agent (Oct 29) that watches the web and posts cited alerts in your channels
The benchmarks (homepage, run Sep 9, 2026) are strong. Quick explainer: benchmarks are standardized test sets everyone runs to compare accuracy and cost. These are Parallel's own runs on 50-100 question samples, so season to taste:
BrowseComp (hard, multi-step research): Parallel 74% vs. Exa 70% and Tavily 66%, at well under half the cost.
SimpleQA Verified (quick facts): Parallel’s cheapest tier hits 94% at ~$2 per 1,000 requests. Tavily needs ~$17 to hit the same 94%, and Exa’s low-cost tier gets 91% at ~$8. Parallel’s pricier tiers reach 97%.
Finance: beats OpenAI's built-in search at 30% lower cost on Parallel's private benchmark.
It's closer on WideSearch (57.6% vs. 55.9%), so it doesn't sweep everything.
The customer receipts are strong:
Nooks cut web search costs 70.5% by switching from built-in model search.
Opendoor compressed HOA research from ~10 minutes of manual work to ~2 minutes of verification, on every property it transacts.
Then there's Index by Parallel (May 2026): a system that pays content owners based on how much their content contributed to an agent's answer. It's a content moat and a legal hedge in one move.
The Team
Parag Agrawal's arc writes itself. IIT Bombay, then a Stanford CS PhD, then more than a decade at Twitter, going from engineer to CTO to CEO. Fired by Musk when the acquisition closed in October 2022, then sued with other execs over $128M in severance. He turned down big exec jobs because, as he told First Round, he couldn't just join someone else's mission.
His line on Sourcery in October: “I only want jobs which I wouldn’t hire myself for.”
Co-founder Travers Nisbet is the commercial half: UVA, McKinsey (including time as a Fellow at the McKinsey Global Institute), then strategy at Chime.
The GTM leadership is some of the most elite in tech:
Graham Moreno, Head of GTM. Joined in March 2026 after leading GTM at Windsurf (President through the Cognition acquisition), with Grafana Labs and MongoDB before that. We had Graham on the pod talking sales mafias and what made MongoDB great, where he called Parallel “the most AI-enabled company I’ve ever been at.” His announcement cited “meteoric revenue growth in the last 6 months.”
Liam Mulcahy, Head of AI Native. Helped scale MongoDB, then spent 7+ years in venture, most recently as GTM Operating Partner at Kleiner Perkins (Parallel’s Series A co-lead). He’s been on the pod twice, most recently on why he left VC to get back in the operating seat.
The cap table is stacked: Sequoia (Andrew Reed on the board), Kleiner (Mamoon Hamid on the board), Khosla, Index, First Round, Spark, and Terrain, with every existing investor increasing in the B.
Culture-wise:
Fully in person (every open role is onsite), with teams in the SF Bay Area, Austin, New York, and London.
A favorite company value: "Try and fail. But succeed an unfair amount."
They ran a Museum of the Human Web pop-up in SF - was very into this and thought it was a cool idea.
Why Join
You can still build the playbook.
Graham and Liam landed this spring, and 30+ roles are open, including two GTM Leaders (AI Native + Enterprise), Enterprise AEs, Technical Enablement, Cloud Partnerships, Product Marketing, and a GTM Recruiter. Agrawal said the B goes toward building out sales and marketing.
The motion is a seller's dream.
A 100K-developer self-serve funnel feeds enterprise deals, Google Cloud distribution lets big companies buy through existing cloud commitments, and per-request pricing means accounts expand as customers' agent volume grows. Sell once, expand with the AI market.
Two deal tempos, both fun.
Per the JDs, AI-native buyers move "at the pace of their release cycle," while enterprise deals are six-to-twelve-month cycles with security review, procurement, and a buying committee.
You'll build real technical chops.
This is a technical sale. You need to hold your own in API and eval conversations with engineers. Pricing pressure is real when search costs about $1 per 1,000 calls, Exa is well-funded with more developers, and the model labs are always a threat. And it's in-person.
The timing math.
Valuation nearly 3x'd in five months, the closest comps just validated the category, and the company is still small. If Parallel becomes the retrieval layer for the agentic web, $2B will look cheap.
If you're interested, reach out or DM me on LinkedIn and I'll make an intro.

Thanks for reading - we'll try to publish these monthly with the top companies we're working with.
For those who are new, my name is Chris Balestras, co-founder & head of talent, media, & brand @ Crew - a GTM recruiting, media, and investing firm, working with seed through Series D AI-natives to help them grow.
Where to find Crew:
We work with many of the hottest AI-native startups in various capacities, and for those who are interested, shoot me an email at [email protected] or a DM on LinkedIn.
🫡 cheers,
Chris
