# How to pass the 8090 interview, Chamath’s AI software factory

Source: https://www.techinterview.org/post/3233477445/8090-interview-guide/
Updated: 2026-09-17 · techinterview.org

8090 hires engineers who can sit in front of a regulated customer, take an ambiguous problem, and ship production software with an audit trail behind it, using AI agents to do most of the typing. The loop, reconstructed from public postings and comparable enterprise-AI shops, runs like a small senior team: a screen, a practical coding round in Python or TypeScript, an architecture round that leans hard on governance and correctness, an agentic build exercise, and a customer-facing conversation. Expect a senior bar and a lot of ambiguity. No current posting lists a salary range and Levels.fyi has no data, so treat any comp number you see online as invented and get the real one from the recruiter.

In June 2026, Chamath Palihapitiya took the CEO title at 8090, his first full-time operating role since Facebook, and announced a $135M Series A led by Salesforce Ventures, per [TechCrunch](https://techcrunch.com/2026/06/29/chamath-palihapitiya-raises-135m-series-a-for-his-ai-coding-startup-takes-ceo-role/). That matters for candidates: a high-profile founder-investor back in a day-to-day operating seat means the company is in build-and-hire mode, and the bar is set by someone with strong opinions about what good looks like.

8090 Solutions Inc. (site at 8090.ai; not Factory AI, not the number) calls its product a Software Factory. The pitch is that AI agents build production-quality software for regulated enterprises, healthcare, insurance, aerospace, energy, financial services, and government, with the controls those buyers require, including a full audit trail of what the agents did and why. The Series A round, alongside Salesforce Ventures, drew WndrCo, Craft Ventures, The Production Board, and Launch, plus angels including Nikesh Arora and Adam D'Angelo, again per [TechCrunch](https://techcrunch.com/2026/06/29/chamath-palihapitiya-raises-135m-series-a-for-his-ai-coding-startup-takes-ceo-role/) and 8090's [own announcement](https://www.businesswire.com/news/home/20260626795833/en/8090-Raises-$135M-Series-A-to-Accelerate-Their-Rollout-of-Software-Factory).

## Why the audit trail changes what they hire for

8090 is easiest to understand by contrast with the tools it isn't. [Cursor](/companies/cursor/) and GitHub Copilot are in-editor assistants: a developer stays in the loop, in an active session, accepting or rejecting suggestions. Lovable and Bolt spin up prototypes fast. 8090 aims elsewhere, at the outsourcing and legacy-modernization budgets of regulated enterprises, coordinating requirements, architecture, agents, and testing across a whole project and leaving a record a compliance officer can read. One [independent analysis](https://rywalker.com/research/8090-software-factory) pegs the enterprise tier as a custom, fully managed engagement starting around $1M a year; treat that as reported color, not a confirmed price.

That target market decides the hiring profile. If the deliverable is software a bank or a hospital will run, generation speed is table stakes. Correctness and traceability are the product, and so is the ability to talk to a nervous stakeholder. The postings that are up read as customer-facing rather than heads-down platform work, which suggests the job is closer to a forward-deployed engineer than to an internal-tools role. Read that as inference from the live listings, not a formal company statement. It resembles the model at companies like [Harvey AI](/companies/harvey-ai-interview-guide/) and [Decagon](/companies/decagon-interview-guide/), where the person shipping the code is also the person in the room with the client.

## Who this is actually for

Calibrate your own level before you apply. 8090 is a small, senior team, around 32 people per Built In as of mid-2026, staffed with engineers and tech leads out of Amazon, Meta, and Google. The application path itself signals the bar: beyond the posted roles, the careers page invites people to email evidence of exceptional ability to hiring@8090.inc, which is not how a company with a wide new-grad funnel talks. One aggregator lists a full-stack role as "0+ years," which is likely a scraping artifact rather than a real junior track, though I can't confirm it either way. I've seen no confirmed early-career or new-grad pipeline; assume the target is mid-level through staff-plus engineers who are comfortable owning ambiguity and facing a customer.

Settle one thing before you invest a prep week: this is an agentic software-delivery and product-engineering role, not an ML modeling or ML-infra job, so a model-research or ML-platform background probably isn't the fit here. The core stack, per public postings, is Python, TypeScript, React, and AWS, with DevOps, data engineering, and hands-on LLM or agentic-AI experience valued. You don't need agents in production on your resume, but you should be able to reason about getting verifiable work out of an unreliable model.

## The loop, reconstructed

8090 is new enough that there are almost no candidate-reported writeups, and the careers page loads its roles dynamically so even the exact titles shift. The table below is not 8090 intel; the stages are modeled on how customer-facing engineering loops run at comparable regulated-enterprise AI shops like Harvey and Decagon, then fitted to 8090's public postings and stated priorities. Read it as the likely shape, not a schedule anyone has confirmed for 8090 specifically.

| 8090 interview stage (reconstructed, not candidate-reported) | Typical format | What it screens for | Plausible prompt |
| --- | --- | --- | --- |
| Recruiter or hiring screen | Phone | Level fit, why enterprise AI, comfort with customer-facing work | Walk me through a project where the requirements were unclear and you shipped anyway. |
| Technical coding round | Live coding in Python or TypeScript | Practical full-stack fluency, clean and correct code, testing instincts | Build an endpoint that ingests messy customer records, validates them, and returns a reconciled view. |
| Architecture and design | Live design discussion | System design with governance, auditability, correctness under regulation | Design a service that lets an auditor replay every change an AI agent made to a codebase. |
| Agentic build exercise | Take-home or live | Getting verifiable work out of an LLM, reviewing agent output, knowing when to distrust it | Use an agent to implement this feature, then show us how you'd prove it's correct. |
| Customer-facing round | Live conversation | Explaining tradeoffs to a non-engineer, scoping ambiguity, stakeholder trust | A client insists on a design you think is wrong. Talk me through the conversation. |
| Behavioral, often with a founder | Live conversation | Ownership, judgment, bar-raising, why 8090 | Tell me about the highest-impact thing you built and what it cost you to get it right. |

A loop at a team this small usually resolves in two to three weeks. Don't read the compact size as an easy bar; a company of about 32 people is picky about its next engineer in a way a 500-person org isn't.

## The design round is really a governance round

The architecture conversation is where the job's premise shows up. A generic system-design answer, add a cache, shard the database, put a queue in the middle, gets you partial credit and no further. What the interviewer is listening for is whether you treat auditability and correctness as first-class requirements rather than features you bolt on later. If AI agents are writing and changing enterprise code, someone has to be able to answer, six months later, what changed, who or what approved it, and whether it still does what a regulator was promised.

So expect prompts about replayable change history, permissioning around what an agent is allowed to touch, test generation and verification, and how you'd contain a bad agent action before it reaches production. The ordinary building blocks still matter, and it's worth an evening with the [standard system design patterns](/system-design-interview-guides/), but here they serve a compliance story. The strongest candidates can name the specific failure mode a regulated customer fears and design directly against it.

## Coding and the agentic exercise

The live coding round is practical, not a puzzle gauntlet. Think one solid full-stack problem: parse and reconcile untidy input, model it cleanly, handle the edge cases, write the tests you'd actually want. Correct and readable beats clever. Your [complexity analysis](/big-o-cheat-sheet/) should be sound, but no one is scoring you on an exotic dynamic-programming trick.

The part that trips people up is the agentic exercise, because the skill it tests is newer. The company's thesis is that agents write most of the code and humans are responsible for it being right. So the interviewer wants to see you drive an agent, then review what it produced, catch where it hallucinated an API or skipped a case, and prove the result works. Treating yourself as the senior engineer reviewing a fast, confident, occasionally wrong junior is the mindset they're hiring for. The broader shift this reflects is worth understanding through the [AI-era interview playbook](/ai-era-interview-guide/).

## The customer-facing round is not a formality

If the roles are as customer-facing as the listings suggest, communication is graded, not assumed. Expect a scenario with a difficult stakeholder or a moving spec, and the interviewer watching how you scope the ambiguity, push back without alienating anyone, and explain a tradeoff to someone who doesn't write code. Have a real story about a project you owned end to end while the requirements moved under you, including the part where it went wrong. Vague ownership claims come apart quickly when the person asking follow-ups sat across from a customer last week. The [STAR-method behavioral prep](/post/3233460379/behavioral-interview-questions-2026-star-method-amazon-leadership-principles-and-winning-answers/) is the right frame; just make the stories specific enough to survive scrutiny.

## Comp, location, and work authorization

There's no public comp data to anchor on here. No current 8090 posting lists a salary range, and Levels.fyi shows nothing for the company, so any figure floating around online is a guess. The numbers only the recruiter has: base for your level and city, share count, the strike or preferred price behind them, and the vest. With those you can price the equity against this Series A yourself. Base should track other well-funded Bay Area startups given the round size and the CEO's profile; the equity is where the real spread sits at a 32-person company. Run competing offers through a [total-comp calculator](/total-comp-calculator/) to compare like with like, and if you reach an offer the standard [negotiation tactics](/post/3233474669/salary-negotiation-2026/) hold, remembering that pre-exit private equity is worth what the next priced round says it is.

Public postings put the roles in the Bay Area (Redwood City / Menlo Park) and in Toronto, and describe them as on-site. No listing I found states a visa-sponsorship policy, so if you need sponsorship or you're eyeing the Canadian office for work-authorization reasons, ask the recruiter on the first call.

## Where your prep time actually pays off

Spend the first hour on the business, not the algorithms: read the Series A coverage and the company's own material, and be able to say in a sentence why a regulated enterprise would pay for agent-built software with an audit trail instead of an in-editor assistant or a traditional consultancy. Then do one concrete drill, and budget an evening for it: open an unfamiliar open-source repo, point a coding agent at it (Cursor, Claude Code, or Copilot Workspace all work), and ask it to add one small feature, a new endpoint or a validation rule. When it hands you a diff, don't ship it: write the test that catches where it faked a nonexistent API or skipped an edge case, then fix what it got wrong. That review-and-verify habit is what the agentic exercise grades. Refresh practical full-stack coding and the design patterns too, and prepare one customer-facing story solid enough to survive a founder's follow-ups. To calibrate the bar, the [AI-startup interview difficulty index](/ai-startup-interview-difficulty-index/) and the wider set of [AI-native company guides](/ai-startup-interview-guides/) help.

The engineers who thrive at 8090 like the uncomfortable middle of this job: half of it is getting a model to do good work, and the other half is being the person a hospital's IT director trusts with the result. If that combination sounds like the fun part of the work rather than the tax on it, this is a company worth the effort of applying well.