B4 Index / The Continuum / August 5, 2026

The Continuum

B4 Research August 5, 2026

Week of July 29 to August 5, 2026 · 3 score movements, 7 capability signals, 35 funding events, 6 papers

1,603

Categories tracked

3

Score movements

$39.6B

Capital tracked

7

Capabilities confirmed

6

Papers reviewed

Order Management (OMS)

Deeper, source-disciplined research showed the prior's '4 / mainstream self-build' rested on weak evidence (hyperscaler self-builds treated as a template, plus off-target OSS/BSS sources). The 2026 reality is that every documented OSS OMS build is hybrid — order orchestration on Medusa/Vendure/Temporal atop existing ERP/WMS and packaged components — and a full distributed-OMS self-build covering ATP, sourcing optimization, allocation, and returns reconciliation is a strategic exception, not mainstream. The market did not move; the prior overstated buildability, so this is an evidence correction to a 3.

AUG 05

Score movements

  • Last-Mile Delivery

    Source-disciplined research showed the prior's '4' anchored on hyperscaler/on-demand self-builds (UPS ORION, Amazon, DoorDash) that are frontier deployments by large infra teams, not a mainstream mid-market template.

  • Business Process Management (BPM)

    Moved 3 to 4 on substantially stronger named production evidence than the prior pass surfaced: Netflix, OpenAI, Vinted, Goldman Sachs, Deutsche Telekom, Zalando, Zoom, and small teams like Nooks all run self-hosted orchestration alternatives in production at scale, making the self-build path a documented mainstream choice rather than a well-resourced-team option.

Funding

35
CompanyRoundCategoriesAnnouncedAmount
OpenAISourceStrategic InvestmentFoundation Model APIs (LLM & Multimodal)JUL 31$35B
AirtableSourceAcquisitionCollaborative Database & No-Code Ops Platform+1AUG 04$1.3B
SemiAnalysisSourceVenture Fund (target)JUL 31$400M
Mariana MineralsSourceSeries BAUG 03$310M
Volta InfraSourceGrowth/EquityGPU Cloud / AI Infrastructure PlatformAUG 04$300M
VoltaSourceSeed + Series AGPU Cloud / AI Infrastructure PlatformAUG 04$300M
MooveSourceSeries CAUG 05$250M
Horizon3SourceSeries EBreach & Attack Simulation (BAS)+1AUG 03$250M
Horizon3.aiSourceSeries EBreach & Attack Simulation (BAS)+1AUG 03$250M
SimileSourceSeries BAI Agent Simulation & Pre-Deployment Testing Platform+1JUL 30$200M
PermisoSourceAcquisitionIdentity & Access Management (IAM)+1JUL 30$200M
HappyRobotSourceSeries CAI Agent Frameworks & Orchestration+2AUG 04$150M
ZenitySourceSeries CAI Agent Identity & Authorization Platform+2AUG 03$125M
DecadeSourceSeedWealth Management PlatformAUG 04$85M
Obsidian SecuritySourceSeries DNon-Human Identity (NHI) Security & Governance+1AUG 04$85M
InforcerSourceSeries CShadow IT Discovery (Lightweight, Non-CASB)JUL 30$50M
inforcerSourceSeries CMulti-Tenant RMM (Remote Monitoring & Management) for MSPsJUL 30$50M
DataBahnSourceSeries BData Pipeline / Ingestion Connector Platform (ELT)+1JUL 30$40M
WindBorne SystemsSourceSeries BAUG 05$37M
AntiochSourceUndisclosedDigital Twin Platform (Operational/Simulation)AUG 04$31.4M
BaselayerSourceSeries AKYB Business VerificationAUG 03$20M
JuneSourcePre-seedAI Agent Frameworks & OrchestrationAUG 03$20M
Bloom SecuritySourceSeedEnterprise Secure Browser+1JUL 30$20M
DiliSourceSeries ARegTech / ComplianceJUL 30$15M
Smallest.aiSourceSeries AVoice AI Platform (Real-Time STT/TTS/Voice Agents)JUL 30$13M
RWXSourceSeries AAI Agent Code-Execution Sandbox Platform+1AUG 04$12M
IntropySourceSeedSpare Parts / MRO Inventory OptimizationJUL 30$11M
Endeavor Optical Networks (EON)SourceSeedAUG 04$10.8M
WorldscapeSourceSeed extensionDefense Command & Control / Battle Management System (C4ISR)AUG 05$10M
EllisSourceSeedPortfolio Company Monitoring (Private Equity)+1JUL 31$10M
Ellis AISourceSeedPortfolio Company Monitoring (Private Equity)JUL 30$10M
Intelligence (Design Arena)SourceSeedAI Model Evaluation (Evals) PlatformAUG 03$7.9M
Actualyze AISourceSeedAI Spend Management & FinOps Platform (LLM/GPU Cost)+1AUG 03$7M
Osmo StudioSourceSeedAI Video Generation PlatformAUG 05$5M
ProfoundSourceSeedAUG 04$1.5M

Capability signals

7
  • BenchmarkedPersistent self-evolving agentic coding runtime (~78% SWE-Bench Pro)

    Evidence AUG 05 · Source

  • BenchmarkedLong-horizon web search agent (BrowseComp 37.3% -> 55.3% with context mgmt)

    Evidence AUG 05 · Source

  • BenchmarkedProactive bug-fixing agent benchmark (1,663 tasks, 8 languages)

    Evidence AUG 05 · Source

  • Production ProvenBuilding a domain foundation model in a data-scarce vertical

    Evidence AUG 05 · Source

  • AnnouncedRunware Sonic Inference Pods — modular containerized inference data centers

    Evidence AUG 04 · Source

  • DemoedNVIDIA Alpamayo 2 Super — open 34B VLA model for L4 autonomous driving

    Evidence AUG 04 · Source

  • DemoedOrchard — open framework for training/evaluating agents at scale

    Evidence AUG 03 · Source

Papers

6
Large Benchmark

Enterprise-grade coding agents executed hidden malicious commands smuggled inside benign-looking skill files in 95.5-96.1% of runs (Gemini CLI) and 71.6-74.0% (Qwen Code), nearly invariant to the generating model, with explicit safety recognition in only 1.99% of runs.

Towards a Risk Assessment of Malicious Skill Files in Coding Agents

What it means Treat third-party agent skill/plugin files as untrusted executable code - sandbox and human-review them before adoption and assume the agent's own safety recognition is near-zero.

arXiv preprint · AUG 05

Large Benchmark

When the issue report is removed and agents must discover as well as fix bugs, most state-of-the-art coding agents struggle - showing limited ability to locate and resolve recorded bugs, handle multiple-bug scenarios, and find valid latent bugs.

Active-SWE: Benchmarking Coding Agents for Proactive Bug Fixing without Issue Reports

What it means Do not extrapolate SWE-bench-style resolve rates to unscoped bug-hunting - agents lean heavily on a human-written issue report to perform; proactive discovery still needs a person to frame the problem.

arXiv preprint · AUG 05

Case Study

Repository-preserved Agent Plan files are rare but informative: screening 36,710 GitHub repositories surfaced only 85 Markdown plan files from 10 repositories, and where present they guided agent execution most often through implementation steps, concrete files and locations, and testing/validation information.

An Exploratory Study of Agent Plans for Agentic AI Coding Tools in Open-Source Software

What it means Preserve agent plan files in-repo with the elements that actually steer execution — numbered implementation steps, concrete file paths, and test/validation criteria — they double as durable human-agent task-intent documentation, but adoption is still near-zero so this is convention to establish, not a settled norm.

arXiv preprint · AUG 05

Case Study

Chat panels, terminal agents, generated diffs, and streaming status output in AI developer tools create real visual accessibility barriers for blind, low-vision, and color-vision-deficient developers, clustering into screen-reader/AT barriers, contrast and differentiation problems, and readability/scaling limits — with prominence varying by tool ecosystem.

Characterizing Visual Accessibility Issues in AI Developer Tools: An Empirical Study

What it means Include screen-reader, contrast, and scaling checks when adopting or building AI coding-tool interfaces — the agent/diff/streaming surfaces introduce accessibility barriers that general IDE accessibility work does not cover.

arXiv preprint · AUG 05

Controlled Study

Prompt wording changes where agent effort is spent without changing success: 'multiple approaches' phrasing inflates reasoning 2.4-7.4x with no success gain, 'maximum certainty' phrasing drives redundant-verification runs costing 18x the clean-run median, and harness design shifts cost per successful task 5-30x.

Same Task, Different Work: Prompt-Induced Waste in Coding Agents

What it means Strip effort-inflating phrases ('explore multiple approaches', 'be absolutely certain') from agent prompts and use bounded-efficiency wording — prompt style is a direct cost lever with no measured correctness payoff.

arXiv preprint · AUG 02

Large Benchmark

13.6% of SWE-bench Verified instances have misaligned PR-issue pairings — the problem statement does not actually match what the graded patch fixes — across five misalignment patterns.

PAIChecker: Uncovering and Checking PR-Issue Misalignment in SWE-Bench-Like Benchmarks

What it means Discount small SWE-bench Verified score differences between models — a double-digit fraction of the instances measure something other than issue resolution, so single-digit leaderboard gaps are within construction noise.

arXiv preprint · JUL 30

Every row here is confirmed before it publishes, and the research is free. The full database and the score updates behind it are in a B4 subscription.