AutogenAI

Somewhere right now, a deadline is winning

It is 11 p.m. and a bid team is still awake. A government tender is due at noon. There are 140 questions, a strict word count, compliance boxes that must all be ticked, and a folder of past answers nobody can find. This scene repeats in thousands of offices every week. AutogenAI exists to delete it.

The company builds custom AI language engines - not a generic chatbot, but a model shaped around one client's voice, evidence and win themes. Feed it the tender, and it qualifies the opportunity, extracts the requirements, drafts persuasive answers, checks them for compliance, and hands a human the part that actually needs a human: judgment. Today more than 200 enterprises run on it, from Fortune 500 firms to construction companies to charities chasing grant money.

Proposals were the last thing left to automate

Every industry got its software. Sales got a CRM. Finance got a ledger that balances itself. Bid writing got - a deadline and a lot of coffee. The work of winning contracts stayed stubbornly manual: brilliant people retyping the same answers, hunting for a case study from two years ago, rewriting a paragraph for the eleventh time because the word limit moved.

It is a strange place for so much money to hide. Public procurement and enterprise bidding move trillions, and the difference between winning and losing often comes down to who could write more, faster, and more precisely under pressure. The process was slow not because the people were slow, but because nobody had built them a better tool.

A bid writer who actually read the philosophy of language

Sean Williams did not arrive at this problem from a lab. He studied philosophy at Cambridge - logic and mathematical logic, of all things - then spent roughly a decade inside government welfare-to-work programmes, eventually running large operations at G4S. He knew the bid room from the inside. He also, inconveniently for anyone who likes a tidy origin story, had already built and sold a training company, Corndel, for around £47 million.

So when generative AI matured, Williams made a specific wager: that the technology was finally good enough to draft persuasive, evidenced prose - if you trained it narrowly, per client, instead of hoping a general model would guess. He founded AutogenAI in late 2021 with co-founder Raj Khaira and shipped the first software by September 2022. The bet was that custom beats generic, and that writers paired with a machine beat writers alone.

Five tools, one engine, zero blank pages

AutogenAI is not one button that spits out a proposal. It is a workflow that walks a bid from "should we even bid?" to "submitted." The thread holding it together is a language engine trained on the client's own content, so the words it drafts sound like the client and cite the client's evidence - not a generic internet average.

The numbers clients put their name to

Claims are cheap in AI. AutogenAI's pitch rests on two figures its customers report: roughly a 30% uplift in win rates - which the company tallies at over $100M of additional work - and up to 85% lower proposal writing costs. Named customers include Serco, HGS and Twin Training, and the buyer base spans Fortune 500 firms, consultancies, construction companies and nonprofits.

Who is backing it

The Series B was co-led by Salesforce Ventures and Spark Capital, with Blossom Capital - an early believer - following on. Salesforce Ventures has also helped open doors into enterprise accounts, the exact buyers AutogenAI needs. The technology stack leans on leading large language models, including Anthropic's Claude.

Augmented, not automated

It would be easy to sell this as "AI replaces the writers." AutogenAI sells the opposite. Its framing is augmented intelligence - the machine drafts, the human decides. The team is deliberately mixed: linguists and writers sitting next to machine-learning engineers, data scientists and UX designers. The point is not fewer people, but people who spend their hours on strategy instead of formatting.

That is also a quietly self-interested philosophy. A proposal that reads like a robot wrote it loses. A proposal that reads like your best writer, produced in a fraction of the time, wins. The mission - AI as a supercharger for human ingenuity - happens to be the same thing as the product spec.

The night shift is ending

Go back to that office at 11 p.m. The tender is still due at noon. But now the requirements are already extracted, the first draft already written in the company's own voice, the compliance boxes already checked. The team is editing, not generating - arguing about strategy, not hunting for a two-year-old paragraph. They go home.

That is the change AutogenAI is selling, and the reason investors put $65.3M behind it. If winning contracts is partly a writing contest, then whoever writes faster and sharper has an edge - and that edge is now buyable. The skeptic's fair question is whether self-reported win-rate gains hold up across industries and over time. The company's answer is to keep shipping, keep its customers, and let the renewal rate do the arguing. Either way, the all-nighter - that small, stubborn ritual of the bid world - is finally optional.