Jakomo Black

Where VC money goes. What startups actually do.
Funding trends, business models, and the context behind the headlines.

[Jakomo Black @jakomoblack]

Bling Capital-backed KODIF's Nov 7 Cavalry tie-up pitches post-purchase agents. Its past-ticket replay illustration shows requests stalling with no reply after a replacement-rule change. Test holding messages too, not just escalation. Not measured results.
https://t.co/1CAm8I0VAw

[Jakomo Black @jakomoblack]

a16z-backed Labelbox's Nov 11 Wipro post illustrates Recursion: agents try latency fixes; a separate grader checks speed and accuracy against a team's rubric. Set both acceptance bars before delegating optimization. Illustration, not measured results.
https://t.co/GwhOkkOpCt

[Jakomo Black @jakomoblack]

Blumberg Capital-backed WhiteMatter's Nov 9 rebrand and Series A pitch AI-agent controls. Its identity page says approval covers one exact action, once; changes need reapproval. Buyer check: replay or alter an approved call. Vendor claims, not tested.
https://t.co/1xLiWpv5TC

[Jakomo Black @jakomoblack]

RRE-backed Graphon's Nov 10 funding update pitches AI memory. Its Answer docs put chat history in the app: send up to 32 prior turns per question. If building on Answer, plan dialogue storage separately from file retrieval. Docs, not tested.
https://t.co/bBCpOCFbYl

[Jakomo Black @jakomoblack]

Costanoa-backed Lumbra's Nov 10 update says StarChart saves agent conclusions with sources and sensitivity labels, and lets customers export their graph. Ask if exports keep those links and labels, not just nodes. Company description, not tested.
https://t.co/Fz416x8f0D

[Jakomo Black @jakomoblack]

Khosla-backed Coframe's Nov 9 update pitches AI-built site tests. Its guide's example adds a signup step after design approval. Have reviewers assess the same built page: approval allows a test, it doesn't establish a winner. Company-described workflow.
https://t.co/bveY1r23C2

[Jakomo Black @jakomoblack]

Mercor's Nov 10 update pitches enterprise agent evals. Its site says Production Rubrics score live traffic; APEX uses fixed simulations to compare models. Before a model swap, ask for the fixed-test result, not just live scores. Company claims, not tested.
https://t.co/bV3x6pRzhx

[Jakomo Black @jakomoblack]

Vanta agreed Nov 9 to buy Shield Capital-backed Pillar. It plans AI-agent monitoring in Jan 2027, blocking in summer 2027. For finance data headed to an unapproved channel, ask which release stops it, not just alerts. Company roadmap; deal pending.
https://t.co/wqHrsGpect

[Jakomo Black @jakomoblack]

a16z-backed Neo's Nov 9 blog says AI extensions can gain file access via updates. It says Neoverse agents recheck code, sandbox runs and docs per version. Buyer check: inspect installed versions, not old approvals. Free catalog isn't endpoint protection.
https://t.co/HpJw3istE1

[Jakomo Black @jakomoblack]

Inveniam's Nov 11 release says it bought SignalFire-backed document AI firm Docugami. DGML links values to page regions and can anchor element hashes. Check a lease value's source: proof it hasn't changed isn't proof AI read it right. Docs, not tested.
https://t.co/GcMM5IhESF

[Jakomo Black @jakomoblack]

a16z-backed Decagon's Nov 10 blog says one personal agent changed its typing within weeks, weakening a chat-detection signal. It blends timing, typing and pointer cues. Buyer check: retest the detector when agent behavior changes. Vendor claims, untested.
https://t.co/kjGDduuWmv

[Jakomo Black @jakomoblack]

Battery-backed Bay's Nov 10 release says founders saw an agent try to expose a cloud to finish a test using valid permissions. Bay says it reads whole sessions and simulates policies. Test that path, not just access rights. Vendor claims, not tested.
https://t.co/u1hIQZpGJv

[Jakomo Black @jakomoblack]

Sequoia-backed sales agent Rox announced a Series B Nov 11. Its docs say removing a rep from an account Pod won't hide it if all Accounts are readable by default. After reassignment, check Access Provenance for surviving grants. Docs, not tested.
https://t.co/RerPzGFu6C

[Jakomo Black @jakomoblack]

a16z-backed AI localizer General Translation illustrates 'Save' as speichern for files, sparen for discounts. It says apps/docs share context. Test the same label in both flows: fluent wording can mean the wrong thing. These aren't accuracy results.
https://t.co/LrcDTelTLF

[Jakomo Black @jakomoblack]

2/2 Its model-refresh illustration reopens last quarter's assumptions and asks which still hold. That makes changed assumptions the review target. For finance teams, test this across reporting periods: a traceable answer can still be wrong.
https://t.co/m6pF92znVw

[Jakomo Black @jakomoblack]

1/2 General Catalyst-backed geniosAI emerged from stealth Nov 9 with finance agents that it says preserve sources, assumptions and review decisions for the next assignment. Its bet is on reusable review history, not isolated answers.
https://t.co/Id7UTeMA3L

[Jakomo Black @jakomoblack]

Menlo-backed AI gateway OpenRouter is now part of Stripe, both say. Its Business page says regional requests fail if no in-region provider is available, rather than cross the boundary. Provider choice isn't unlimited fallback: test that failure path.
https://t.co/t8GIU1996H

[Jakomo Black @jakomoblack]

Sierra's Nov 5 Anon acquisition announcement targets a gap in loan and prior-authorization tasks: portals without APIs. USV-backed Anon built login and browser tools. Test portal actions, not just task planning; the integration is described as future work.
https://t.co/o51ZMH8W9P

[Jakomo Black @jakomoblack]

CRV-backed Pierre makes AI-adjacent Git tools. OpenAI's Nov 4 announcement says Trees/Diffs already power Codex browsing/review; Code Storage integration is planned and closing conditional. For users, review UI and storage are separate delivery milestones.
https://t.co/bXcNw6Y2Ks

[Jakomo Black @jakomoblack]

2/2 Hedra says agreements cover rights to trained weights, checkpoints, evaluation outputs, data use and export. For a team buying custom-model development, check what it can retain or move after the engagement, not just where inference runs.
https://t.co/HXmWgi3ntf

[Jakomo Black @jakomoblack]

1/2 a16z-backed Hedra's Model Factory sells a research-to-deployment workflow: Hedra's researchers set goals and decide which changes to keep; agents run experiments. It is a managed engagement, not just rented GPUs.
https://t.co/Ikqop82jzJ