The Agentic Analyst: Why We Invested in Pinegap

meet an equity research analyst at a long/short fund.

They are a sector specialist covering 25 names, each one carrying a carefully built thesis, a set of KPIs they watch obsessively, and a milestone map of what would make them more - or less - convinced.

It is earnings season. Six of their companies report this week, some on the same day. For each one, preparation means two to three hours of pulling consensus revisions, prior-quarter reactions, guidance language, and intra-quarter data points across Bloomberg, Visible Alpha, and CapIQ into a preview note. Before 10 AM, over a hundred sell-side emails have already hit their inbox. Somewhere in that pile is the one datapoint that matters to their thesis - if they can find it.

They are among the most analytically rigorous, intellectually competitive professionals in any industry. And a remarkable share of their week is spent on work they would happily hand to an intern: collecting, deduplicating, reformatting, synthesizing.

This is not an outlier. It is the daily reality of buy-side research - expensive, capacity-constrained analysts, each covering 20–40 stocks and responsible for supporting Portfolio Managers that manage hundreds of millions of dollars in AUM, drowning not in a shortage of information but in an excess of it.

The idea of using software to help is, of course, not new. Terminals like Bloomberg and FactSet have owned this desk for decades, and have begun bolting AI features onto their architectures. A newer generation of AI-native research platforms lets analysts search and summarize vast document libraries. And every analyst today has a ChatGPT or Claude tab open.

Yet all of these share one design assumption: the analyst asks, the tool answers. They are pull-based systems. The analyst still has to know what to look for, remember to look, and reassemble the output into the format their fund actually uses. The recurring workflows that consume their week - the earnings preview built the way their fund builds it, the morning summary filtered through their thesis, the thesis milestone that quietly got breached on slide 47 of a conference deck - remain hers to do manually.

That last mile is unsolved.

this is where Pinegap steps in.

Pinegap builds AI agents that behave like a junior analyst for fund teams. Their pitch to a customer is disarmingly simple: what would you hand to an intern so you could scale yourself? That work is what they have "agent-ified."

The mechanics matter here. Every deployment starts by ingesting the analyst's actual context - their watchlist, their thesis per name with bull case, bear case, and milestones, and the specific KPIs they track for each company. On top of this sit fifteen-odd standard agents - earnings previews, post-earnings recaps, a 24-hour news summary, thesis trackers, a screener that handles qualitative criteria alongside quantitative ones. And then come the custom agents, typically built within 48 hours of an analyst describing a workflow: a primer reorganized around one fund's ten proprietary investment principles; a deal-analysis agent for an event-driven investor that fires automatically on merger announcements and digests a 350-page filing before the market opens.

The critical design choice is that all of it is push, not pull. Outputs land in the analyst's inbox - the news summary at 7:30 AM, the earnings preview five days before the print, the alert the moment a filing drops. Increasingly, their morning starts with Pinegap, not with a terminal. Across their customer base, Pinegap's agents now generate over 50,000 reports a month.

We will be honest: we came in sceptical for various reasons. This is a customer who is very busy and famously hard to impress, the underlying data is largely public, and it is entirely legitimate to imagine a world where foundation models plus data connectors let any fund assemble something similar on its own.

what changed our minds was the customers.

In call after call, the feedback was consistent: this looks and feels like a product built by someone who has done the job. We spoke with analysts who were consolidating spend from far more expensive generic platforms into Pinegap. We heard from funds with serious in-house AI capability who still preferred Pinegap for these workflows. We came to see that the hard part is not the intelligence layer - it is the workflow layer: knowing which questions to ask, in what sequence, calibrated to how a specific analyst at a specific fund actually works. That is domain knowledge, earned one customer at a time.

it helps that one of the founders spent fifteen years earning it.

Ankit Varmani spent seven years on the sell side at JPMorgan and eight on the buy side at a long/short fund - he has lived the exact workflow Pinegap now automates, and it shows in every screen of the product. Deepak Sharma, an engineer who built ADAS systems at Jaguar Land Rover, is the other half: twelve years Ankit's junior, from an IIT BHU batch a decade apart, connected originally through their families.

Before Pinegap had a name, the two spent months running weekly experiments together - Ankit explaining how research works, Deepak building against it every Friday. That rhythm never stopped. Deepak has absorbed the domain at a pace that startled us, and still spends hours every day with customers, hand-crafting their agents until the product fits their process exactly. It is rare to find a team where domain depth and engineering velocity are this complementary - and this obsessed with the same customer.

Within a year of commercial launch, Pinegap counts 100+ institutional funds as customers, has 1000+ agents deployed and is generating upwards of 50,000 personalized reports every month.

We believe research in the world of AI will move from search to push - from analysts hunting for insight to insight finding analysts - and that Pinegap will lead that shift. We are delighted to partner with Deepak and Ankit on the journey.

‍

Read the press coverage here.