Antidote.

Tech Diligence

AI diligence,
evidenced.

Antidote tests economically material software workflows, measures what models and systems can actually do, and keeps capability, cost, defensibility, and unknowns on separate planes.

Investment decisions are moving faster than their evidence.

A model can pass a public benchmark and still fail the workflow. A product can show a lift and still lack a defensible asset. Vendor directories and composite leaderboards are not enough. Antidote keeps denominators, confidence, and claim status visible, and it will not invent a moat score to close the gap.

How a public analysis is built

The same sequence as an engagement. The pages below stop where the evidence stops.

  1. Freeze

    The workflow, cases, and scoring rules lock before the test.

  2. Compare

    Feasible comparisons run. Missing ones stay visible.

  3. Separate

    Capability, cost, defensibility, and unknowns do not mix.

  4. Label

    Every claim keeps a status. A failed test stays open.

Three kinds of work

The client question selects the kind of memo. Capability, cost, and defensibility stay separate in each of them. Published benchmarks attach to the job, not the vendor. Two competing products in the same category share those benchmarks and differ in their company-specific findings.

What an engagement examines

Decision-relevant facts first. Provenance, configuration, and caveats stay expandable rather than hidden.

Current workflow

The incumbent process, included only when it was actually observed.

Generic model

A generally available model with no product-specific setup.

Named competitor

A declared alternative with the same job, tools, and budget.

This product

The system under review, tested with authorized access.

Isolated changes

Included only when access and rights permit. Missing comparisons stay visible.

Products under the method

Diligence memos on software that already sits in buyer workflows, including products that are not model surfaces. Same planes, same comparison conditions, same claim labels. Public numbers stay cited to their source. These pages are not client work.

Market intelligence

Filings, earnings, broker research, and expert-call transcripts. Public filings are exposed to a generic retriever. Permissioned transcripts are the scarce-input hypothesis, not a verified flywheel.

Diligence research agent

Agentic commercial-diligence research over open-web sources with slide output. A June 2026 third-party assessment reports a breadth lead but measures artifact properties, not whether cited claims are correct.

Virtual data room

Permissioned deal rooms, Q&A, and buyer administration. The seller owns the documents. Mid-deal switching can bind; that friction decays after close.

Legal AI

Contract analysis, diligence, and litigation assistance. Client matters are the firm's. Post-training is back on the register; Tenet is a declared research preview, not a measured increment.

Document analysis

Multi-document analysis for diligence and research. The data room is the customer's. Orchestration is reconstructable in kind; decay versus long-context models is open.

Legal DMS

Matter work product, email, and document management inside the firm. The corpus is the firm's. Microsoft 365 is the bundling and AI-invasion surface.

Deck generation

Brief plus source pack to an editable deck. In a July 2026 third-party assessment it leads a frontier add-in on production mechanics while the measure that decides partner-readiness, claim-to-source support, would not stabilize. The incumbent bundle failed the end-to-end job at screening.

Private-market data

A licensed private-market graph for company, deal, investor, and comparable screening. Unpublished facts are the structural hypothesis. They are not a verified flywheel.

Excel modeling agent

Natural-language model construction and checks inside Excel. Official policies establish the product surface and data terms; a named practitioner test finds useful narrow-task output and consequential modeling failures. No independent product test has been run.

HCM and finance system of record

Hire-to-pay workforce administration and adjacent finance on one cloud platform. The employer owns the records. Direct AI substitution of payroll and the org chart fails; Sana and Illuminate-era agents are an overlay. Oracle, SAP, and Microsoft are the named invasion surface.