B4 Research August 12, 2026

The Continuum

Week of August 5 to August 12, 2026 · 1 score movements, 3 capability signals, 13 funding events, 3 new categories

1,603

Categories tracked

Score movements

$11.7B

Capital tracked

Capabilities confirmed

Papers reviewed

Work Management

Deeper research separated the category core from its neighbors: the prior 4 leaned on orchestration-layer builds (Airflow, Temporal, Maestro) and a self-hosted task layer, but 2026 evidence shows the business-user work-management envelope — grids, portfolio hierarchy, resource planning, governance — is bought, with self-builds confined to narrow layers around it. A research correction, not a market move.

AUG 11

Funding

Company	Round	Categories	Announced	Amount
Sony / TSMC image-sensor JV (Advanced Vision Semiconductor Manufacturing Corp)Source	Joint venture investment		AUG 11	$4.7B
Thrive HoldingsSource	Growth		AUG 12	$2B
Thrive HoldingsSource	growth (equity financing round)		AUG 12	$2B
River AISource	Seed + Series A (combined)	AI Reinforcement Fine-Tuning & Post-Training Platform (RFT/RLHF/DPO)+3	AUG 11	$1.1B
Layer GlobalSource	Fund first close		AUG 11	$1.1B
LovableSource	Series C	Design-to-Code Generation+2	AUG 12	$400M
CodeRabbitSource	Series C	Code Quality Analysis	AUG 12	$143M
Skan AISource	Series C	Business Process Management (BPM)+1	AUG 12	$63M
Alpha ComputeSource	Asset purchase (land + gas rights)	Data Center Infrastructure Management (DCIM)+1	AUG 11	$55M
BlacksmithSource	Series B	CI/CD	AUG 12	$45M
MindgardSource	Series A	AI / LLM Security (Runtime Guardrails & AI-SPM)+1	AUG 12	$30M
ClearJetSource	Series B	Air Cargo Booking & Management+2	AUG 12	$25M
ManusSource	M&A (deal unwind / return to independent company)	AI Agent Frameworks & Orchestration	AUG 11	Undisclosed

Capability signals

BenchmarkedSecurity agents that prove exploitability and drive fixes
Evidence AUG 12 · Source
BenchmarkedBrowser/web-navigation agents at ~88% (WebVoyager, per aggregator snapshot)
Evidence AUG 12 · Source
BenchmarkedGrok 4.6: frontier long-horizon agent + coding model
Evidence AUG 12 · Source

New categories

Agent Memory Systems . Persistent/structured agent memory is graduating from research into shipped products and a distinct buy-vs-build decision with no dedicated shelf. Signals this window — GitHub Copilot Memory for JetBrains (Aug 11); Mastra built-in memory (Aug 11); AgentCore Memory to GovCloud (Aug 7); HiGram hierarchical graph-memory research (Aug 5); plus an established OSS field (Mem0, Cognee, Graphiti/Zep, Letta, Hindsight) being actively benchmarked. The decision 'how does my agent remember across sessions' now has its own vendor cluster.
Autonomous Security Testing / AI Vulnerability Research . A capability-graduation cluster landed in one week: OpenAI's Astra reaching preliminary Critical cyber threshold (autonomous zero-day discovery + end-to-end attack planning), the productized Daybreak Red/Blue cyber models, their availability on Amazon Bedrock, and third-party cyber evaluations. Existing shelves (Vulnerability Management, Penetration Testing) predate autonomous AI-driven exploit research and may not capture the AI-red-team/AI-vuln-research decision.
Agent Payments / Autonomous Transaction Rails . Managed rails for autonomous agents to make verifiable, auditable payments are graduating to production (Solv Labs on Bedrock AgentCore payments), alongside broader AgentCore runtime GA. The decision 'how do our agents transact money with an auditable trail' has no dedicated shelf in the current taxonomy.

Every row here cleared the pipeline's verification before it published, and the research is free. The full database and the score updates behind it are in a B4 subscription.