Man Group ships its own multi-model agent harness
Man Group released an in-house harness that connects LLMs to its tools, memory and execution environments so agents can act.
Sourced examples from real firms, sorted by the job they do. Every card shows where the claim came from and who made it. Every use case was reviewed by an editor before it went up; vendor claim means the source is the vendor itself.
Man Group released an in-house harness that connects LLMs to its tools, memory and execution environments so agents can act.
A comparison of what leading funds disclose: Man's AlphaGPT proposes, codes and evaluates signals; Bridgewater's Pocket Analyst Tool serves hundreds of investors with validated dataframe outputs; Balyasny evaluates models inside the full agent environment, not in isolation.
Guinness Ventures' head of technology describes using AI for filtering, research, benchmarking and drafting, with every output checked and investment committee decisions kept human because the models are 'frequently wrong'.
Hanover Park publishes a downloadable skill: ask Claude to 'Draft our Q2 LP letter' and it pulls performance, capital activity and portfolio developments through a read-only connector, adding a tie-out appendix.
Falcon Investment Advisors uses a Claude skill in Cowork to turn 70-page CIMs into a templated financial summary.
Capital Intelligence combines eVestment allocator data with a GP's own CRM to enrich investor relationships and surface new prospects for fundraising.
Unveiled at TechCrunch Disrupt, Arya spans fund accounting, investor servicing, compliance, reporting and capital activity.
Primary's job post sketches the emerging role: one leader owning AI systems for sourcing, diligence, portfolio support and firm operations.
Hg, which runs a 100+ person AI engineering group (Hg Catalyst), is expanding its Anthropic collaboration to deploy agentic capabilities across its software portfolio companies.
Boston PE firm Ethos Capital spent about five years building Petra, an assistant trained on 50,000+ sources including every deck the firm has done or passed on.
Managing partner Fadi Chehadé says Petra answered roughly 90% of his questions on a target's legal risks in minutes, work that would have gone to an outside-counsel associate for weeks.
An open-source Claude and Codex skill for earnings season that reviews the last four calls, cross-checks management claims against filings and outputs an HTML report.
Mr. Family Office sketches what an LLM-enabled office looks like in practice: document intake, internal diligence notes and a searchable knowledge base, with governance and privacy as the gating questions.
Ken Griffin says an internal Citadel system reads finance papers, reimplements the methodology, validates the published results and tests them out of sample in two to three hours, work that took researchers six to eight weeks.
Marc Palet shares the stack he uses for knowledge-base upkeep, research and reporting, and why recurring operational tasks now go to agents rather than a junior hire.
Secondaries buyers use AI to turn heterogeneous LP data rooms into source-linked portfolio views, reconstruct reference-date NAV and analyze every underlying holding.
An oversight agent reviews fund-admin deliverables for GPs, flagging discrepancies before reports go to LPs.
AngelList's MCP connector gives GPs read-only access to their fund records from Claude, ChatGPT and other MCP tools, so questions like 'what's my remaining reserve' get answered in chat.
Five KV talent partners support searches across hundreds of portfolio companies at once.
The Seattle real estate investor built a proprietary platform that reads offering memoranda, rent rolls and T-12s into its underwriting model.
A roundup including Correlation Ventures, whose predictive model trained on 100,000+ financings lets it make co-investment decisions in about 48 hours.
Allvue lays out agents coordinating each step of a capital call (allocation math, notice generation, LP communications, remittance tracking) across GP and administrator teams, with approvals kept human at each gate.
Masterworks trains models on 1.2M+ auction lots, 130,000 artists and 6,000+ auction houses to find signals of future appreciation, such as gallery representation, museum holdings and collector base, then pulls comparables to set a fair-value anchor.
Singapore's Temasek is applying AI to its own investment decisions while pushing adoption across portfolio companies, alongside a plan to raise AI exposure to as much as 15% of the portfolio by 2031.