
Venture capitalAbout 40 peopleSydneymseq.vc
How Main Sequence's team ran its AI skills 592 times in one week
Main Sequence is a deep-tech venture fund founded by CSIRO in 2017, with over $1bn under management.

At a glance
- 592
- Skill runs by the team in one week, up from 473 the week before
- 70,749
- Files from six shared drives, in one AI brain
- 18 builds
- Mapped with five of the team in the first interviews
“I can see very quickly we're going to have like 5 or 6 skills that people will be checking separately, each one, just to see if each of it works.”
Before
The fund's knowledge was spread across six Google Drives, Notion and its CRM. Partners were building AI tools by hand and spending their time fixing them. Answering a simple investor question, like which companies have a defence angle, meant digging through all of it.
What we built
We helped Main Sequence build one AI brain that knows the fund, plus a library of AI skills the team shares.
- Step 1
Every night the brain reads their six Google Drives, Notion and the CRM.
- Step 2
It copies the permissions on each file, so each person only sees what they could already see.
- Step 3
Skills for the team's regular jobs, like valuation prep, live in one shared library.
- Step 4
Two of their team manage the library, so new skills get checked before everyone uses them.
After
In the week to 23 October the team ran its skills 592 times, up from 473 the week before. The brain has read all six drives since August 2026. During the quarterly valuation cycle, the team leaned on a skill one of their own people built.
The full story
A fund that runs on documents
CSIRO founded Main Sequence in 2017 to turn Australian research into companies. Its first fund was $240 million, its second $310 million and its third $450 million, which took it past $1bn under management.
Every company it backs sends board packs, updates and valuations. That knowledge sat across six Google Drives, Notion and the fund's CRM, and finding one answer meant searching all of them.
Starting with the people doing the work
In April 2026 we sat down with five of the team and asked what slowed them down. Together we mapped 18 builds.
For the quarterly valuations, an agent that pulls commentary from email, Slack and board packs into the valuation documents. For due diligence, a review tool that ranks the issues in a bundle of contracts. For deal prep, a search across the portfolio's deal documents for standard terms and check sizes. And a design agent that puts every deck into Main Sequence branding.
One brain, with the same permissions as the drives
Tools built one at a time each needed someone to fix them when they broke. So the centre of the work became one brain the whole fund could point its AI at.
Every night it reads the six shared drives, Notion and the CRM. It copies the permissions on each file, so a person only sees what they could already open in the drive. All six drives were in by August 2026.
A library the team looks after
Skills for the fund's regular jobs live in one shared library. Two of the Main Sequence team are its admins, so a new skill gets checked before everyone uses it, and nobody has to route every change through one partner.
In the quarterly valuation cycle, the team leaned on a skill one of their own people built.
The problem it was built for
On a scoping call in September, Phil Morle, a partner, described the risk of every person building their own tools, each one checked on its own:
“I can see very quickly we're going to have like 5 or 6 skills that people will be checking separately, each one, just to see if each of it works.”










