A live trace of the Noma agent patch protocol against
examples/agent-stale-memo/memo.noma. Five block-level operations
refresh stale citations, raise a claim's confidence, attach fresh evidence,
and escalate one risk. Every unrelated byte stays put.
WARNING [stale-citation] input:32:1: Citation "ambience-q4-disclosure" was last accessed 2025-12-15 (>60 days ago). WARNING [stale-citation] input:48:1: Citation "legal-tech-survey-q3" was last accessed 2025-10-04 (>60 days ago).
1. update_attribute id=ambience-q4-disclosure accessed="2026-05-10" 2. update_attribute id=legal-tech-survey-q3 accessed="2026-05-10" 3. update_attribute id=claim-clinical-leaders confidence=0.74 4. add_block parent=thesis <inline content> 5. update_attribute id=risk-mcp-integration-collapse severity="high"
patchSource — line range onlypatchSource(source, ops) rewrites only the targeted line range
for each op. Frontmatter quoting, sibling blocks, attribute order on
unchanged lines, and blank-line padding all survive byte-for-byte.
---
title: Vertical AI — Q1 2026 Investment Memo
author: ferax564
date: 2026-01-30
tags: [memo, research, vertical-ai]
profile: research
stale_citation_days: 60
---
# Vertical AI — Q1 2026 Investment Memo
::summary
Position paper on the three vertical-AI categories worth tracking heading into
Q2: clinical documentation, legal review, and financial-statement audit. Each
section carries claims, supporting evidence, and a stale-evidence flag that an
agent can refresh on a schedule without rewriting the memo.
::
## Thesis
::claim{id="claim-clinical-leaders" confidence=0.62}
Clinical-documentation agents are 18 months ahead of legal-tech competitors on
production deployment depth. The deployment-rate delta is the leading indicator.
::
::evidence{for="claim-clinical-leaders" source="ambience-2025-q4-disclosure"}
Ambience reported 41 hospital-network deployments at the close of Q4 2025 and
projected 60 by mid-2026. Comparable numbers from leading legal-review vendors
sit in the high-single-digits.
::
::citation{id="ambience-q4-disclosure" url="https://example.com/ambience-q4-2025" accessed="2025-12-15"}
Ambience Q4 2025 transparency disclosure, accessed December 15, 2025.
::
::claim{id="claim-legal-mcp-risk" confidence=0.58}
Legal-review vendors face the steepest disruption from MCP-style adapters
because their workflow value is concentrated in document ingestion rather than
in regulated workflow ownership.
::
::evidence{for="claim-legal-mcp-risk" source="legal-tech-survey-2025-q3"}
Q3 2025 procurement survey: 62% of GCs cited "ingestion + retrieval" as the
single biggest reason they bought a legal-tech product. Workflow ownership
ranked fourth.
::
::citation{id="legal-tech-survey-q3" url="https://example.com/legaltech-survey-2025-q3" accessed="2025-10-04"}
Legal-tech procurement survey, Q3 2025 wave, accessed October 4, 2025.
::
## Risks
::risk{id="risk-frontier-leap" severity="medium" owner="ferax564"}
A frontier-model capability leap (GPT-6-class long-horizon tool use) could
collapse the workflow gap. Most vulnerable: verticals where workflow
complexity comes from reasoning chains rather than data integration.
::
::risk{id="risk-mcp-integration-collapse" severity="medium" owner="ferax564"}
Universal MCP adapters compress the integration moat from weeks to hours.
Watch the EHR + practice-management adapter rollout as the leading indicator.
::
## Quarterly review task
::agent_task{id="quarterly-review"}
Every quarter, walk this memo and:
1. For each `::claim`, check whether new public evidence supports or
contradicts. Add a fresh `::evidence` block and adjust `confidence`.
2. For each `::citation`, check whether `accessed` is older than 60 days
(the `stale_citation_days` window declared in frontmatter). If yes,
update the `accessed` date or replace with a newer source.
3. For each `::risk`, check whether the leading indicators have moved.
Adjust `severity` if warranted. Do not delete prior risks.
Issue patches as block-level operations (`update_attribute`, `add_block`,
`replace_block`) rather than rewriting the memo. The unrelated 95% of
the file should be byte-identical after the refresh.
::
> The point of this memo is not the prose. It is that an agent can
> re-open it next month, walk the claim/evidence/risk graph, and update
> only the parts that changed — leaving the rest alone, producing a Git
> diff that shows exactly which beliefs moved.
---
title: Vertical AI — Q1 2026 Investment Memo
author: ferax564
date: 2026-01-30
tags: [memo, research, vertical-ai]
profile: research
stale_citation_days: 60
---
# Vertical AI — Q1 2026 Investment Memo
::summary
Position paper on the three vertical-AI categories worth tracking heading into
Q2: clinical documentation, legal review, and financial-statement audit. Each
section carries claims, supporting evidence, and a stale-evidence flag that an
agent can refresh on a schedule without rewriting the memo.
::
## Thesis
::claim{id="claim-clinical-leaders" confidence=0.74}
Clinical-documentation agents are 18 months ahead of legal-tech competitors on
production deployment depth. The deployment-rate delta is the leading indicator.
::
::evidence{for="claim-clinical-leaders" source="ambience-2025-q4-disclosure"}
Ambience reported 41 hospital-network deployments at the close of Q4 2025 and
projected 60 by mid-2026. Comparable numbers from leading legal-review vendors
sit in the high-single-digits.
::
::citation{id="ambience-q4-disclosure" url="https://example.com/ambience-q4-2025" accessed="2026-05-10"}
Ambience Q4 2025 transparency disclosure, accessed December 15, 2025.
::
::claim{id="claim-legal-mcp-risk" confidence=0.58}
Legal-review vendors face the steepest disruption from MCP-style adapters
because their workflow value is concentrated in document ingestion rather than
in regulated workflow ownership.
::
::evidence{for="claim-legal-mcp-risk" source="legal-tech-survey-2025-q3"}
Q3 2025 procurement survey: 62% of GCs cited "ingestion + retrieval" as the
single biggest reason they bought a legal-tech product. Workflow ownership
ranked fourth.
::
::citation{id="legal-tech-survey-q3" url="https://example.com/legaltech-survey-2025-q3" accessed="2026-05-10"}
Legal-tech procurement survey, Q3 2025 wave, accessed October 4, 2025.
::
::evidence{for="claim-clinical-leaders" source="abridge-2026-q1-disclosure"}
Abridge reported 73 hospital-network deployments in Q1 2026 — up from 52 at the
end of 2025. The category leaders are now growing deployments at ~40% per
quarter, which is the data point that pushed the confidence revision.
::
## Risks
::risk{id="risk-frontier-leap" severity="medium" owner="ferax564"}
A frontier-model capability leap (GPT-6-class long-horizon tool use) could
collapse the workflow gap. Most vulnerable: verticals where workflow
complexity comes from reasoning chains rather than data integration.
::
::risk{id="risk-mcp-integration-collapse" severity="high" owner="ferax564"}
Universal MCP adapters compress the integration moat from weeks to hours.
Watch the EHR + practice-management adapter rollout as the leading indicator.
::
## Quarterly review task
::agent_task{id="quarterly-review"}
Every quarter, walk this memo and:
1. For each `::claim`, check whether new public evidence supports or
contradicts. Add a fresh `::evidence` block and adjust `confidence`.
2. For each `::citation`, check whether `accessed` is older than 60 days
(the `stale_citation_days` window declared in frontmatter). If yes,
update the `accessed` date or replace with a newer source.
3. For each `::risk`, check whether the leading indicators have moved.
Adjust `severity` if warranted. Do not delete prior risks.
Issue patches as block-level operations (`update_attribute`, `add_block`,
`replace_block`) rather than rewriting the memo. The unrelated 95% of
the file should be byte-identical after the refresh.
::
> The point of this memo is not the prose. It is that an agent can
> re-open it next month, walk the claim/evidence/risk graph, and update
> only the parts that changed — leaving the rest alone, producing a Git
> diff that shows exactly which beliefs moved.
(no issues — clean validate)