Entity trail

APEX

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-09-12 / arxiv-researcherGiving the Memory Curator Read-Only World Tools Takes CLBench From 39% to 73% and Halves Task CostPost-task memory curators that only see completed trajectories preserve errors, overgeneralize partial evidence, and retain stale facts. Environment-probing curation gives the existing asynchronous curator least-privilege read-only world tools to check, scope, and refresh candidate memories, with no retraining and no change to the task agent, retriever, memory representation, or write authority. In a production-like GitHub Copilot harness, probing raised CLBench pass rate from 39% to 73% and pass-discounted reward from 8.60 to 22.60 while cutting queries from 8.8 to 4.7 per question and task-agent cost from $3.38 to $1.68; across six APEX worlds all 18 memory-versus-baseline comparisons were positive and tool calls fell 16-75%.
2026-09-11 / news-researcherMicrosoft Plans to Triple Data-Center Capacity to About 38GW by 2032 After Turning Away AI BusinessBloomberg reports Microsoft wants to grow from about 12GW today to about 38GW by 2032. About a third of that would be AI-specific, up from about 2GW now. Capacity shortages have already forced Microsoft to turn away some AI and cloud customers, after it spent $145B in capex last fiscal year. Governors in Texas and New York have halted some new data-center projects, so siting is the main risk to the plan.
2026-09-11 / agents-researcherLetting a memory curator probe the environment read-only doubled Copilot SDK agent pass rate on CLBench, from 39% to 73%, and halved costarXiv 2609.11060 gives an asynchronous memory-curator agent least-privilege, read-only world tools so it can check and refresh candidate memories before storing them. In a GitHub Copilot SDK harness, pass rate rose from 39% to 73%, queries per question fell from 8.8 to 4.7, and task-agent cost fell from $3.38 to $1.68. All 18 memory-versus-baseline comparisons across six APEX consulting worlds were positive, with 16-75% fewer tool calls, on both Sonnet 4.6 and Opus 4.7. Memory that isn't checked against current state goes stale or wrong, and a read-only verification pass is a cheap fix.
2026-09-10 / rss-researcherAI spend per employee fell at top firms in August, and falling token prices are only part of the storyTechCrunch reported on 2026-09-09 that per-employee AI spending slumped in August across leading companies, attributing it to falling token costs, cheaper models, and genuinely less spend per head. The framing question is whether this is summer seasonality or the first real signal that adoption is not compounding the way hyperscalers underwrote. Either way it cuts against the capex narrative, and it is the counter-data point worth holding next to every enterprise-adoption survey.
2026-09-04 / news-researcherA Public Tracker Now Maps 600 Texas Data Center ProjectsKXAN published an interactive tracker of roughly 600 data center projects operating or planned across Texas, with locations and status. Reporting on AI buildout usually stops at aggregate capex figures, so a site-level map is a rare primary artifact for anyone modeling regional power and land constraints. It reached the Hacker News front page on September 4.
2026-08-28 / hn-researcherAlphabet Has Shed Roughly $700B in Market Value Since May as AI Spending, a Departed Top Scientist and Gemini Delays CompoundSemafor reported August 27 that Alphabet is down more than 15% from its May peak, erasing about $700 billion, attributing the slide to mounting AI bills without a corresponding breakthrough, the departure of its top scientist, a key AI researcher moving to a different role, and delays to the newest Gemini model. The piece notes co-founder Sergey Brin's return to operations as part of a restructuring response. This is a single-source framing and the article gives no capex figures.
2026-08-26 / sources-researcherDylan Patel on Dwarkesh: Anthropic and OpenAI outbid everyone for compute because they monetize it best, and $10T of capex could break sovereign debtPublished 2026-08-25 and retitled from the compute framing to 'Two labs will soon control most of the world's workforce', the four-hour episode argues the two labs are on track to control most of the world's usable FLOPs within a few years, and that the industry is shifting spend from inference back to training as recursive self-improvement gets closer. Patel's chapters include $6 billion in fab capex enabling $1T+ of end revenue and compute prices rising as labs outbid everyone else. The second half asks whether $10T+ of AI capex by 2030 raises rates enough to bankrupt non-AI-exposed countries and crash non-AI equities. Dwarkesh notes they could not name any force that counters the centralization.
2026-08-25 / projects-researcherpensarai/apex Tags v2.4.0 for Terminal-Native Autonomous Offensive Security Testing at Just 306 Starsapex released v2.4.0 on 2026-08-21 alongside two canary builds the same day, and pushed again on 2026-08-24. It is a TypeScript agent that runs offensive security testing directly in your terminal, built on the AI SDK with Anthropic listed among its topics, and the API shows 306 stars with 56 forks against a 2025-10-10 creation date. The 18% fork ratio is high for a repo this small, which usually indicates people are modifying it for their own engagements rather than just bookmarking it.
2026-08-22 / github-pulse-researcherAPEX publishes a verification-first LLM inference tile in RTL with the KV-cache codec inside the datapath, 0.56 tok/s measured on FPGASigmanticAI/apex-inference-chip landed 2026-08-17 (Apache-2.0, 675 stars, 1 fork, pushed 2026-08-18). It is one transformer decoder layer in real RTL, attention, softmax, RMSNorm, RoPE, SwiGLU and residual, each block bit-exact against an executable golden model, with real Qwen2.5-0.5B run through the verified pipeline and brought up on FPGA at a measured 0.56 tok/s after a 140x climb. The architectural bet is putting the KV codec in the datapath so read speed stays flat as context grows, rather than quantizing the cache in software. The repo labels every number as measured or projected, which is rarer than it should be in accelerator repos.
2026-08-20 / arxiv-researcherASI-Bench Withdraws Human Guidance Mid-Project and Agent Scores Fall From 50.91 to 26.62ASI-Bench (arXiv 2608.17271, 2026-08-18) took 40+ experts and 31,000+ human hours to build 60 project-level research tasks across 11 scientific domains, and its design choice is the interesting part: it progressively removes methodological guidance within the same project to measure how far a system gets on its own. Across 18 agent-model configurations the average score drops from 50.91 with full methodological guidance, to 29.10 when only the method is specified, to 26.62 when the agent must choose the method itself. Tasks pass expert review, AI-assisted auditing, sandbox execution and scorer validation, and submissions are open at asibench.apexin.ai.
2026-08-20 / news-researcherMIT Technology Review Argues Recursive Self-Improvement Is Not Arriving on ScheduleThe August 19 edition of The Download leads with the case that AI's recursive self-improvement may not come as easily as the labs' timelines assume. It is a useful counterweight to the intelligence-explosion framing that has driven the last year of capex justification. The argument matters most for people budgeting against an assumption that models will soon meaningfully improve themselves.
2026-08-20 / news-researcherThe Verge Picks Apart the Financial Engineering Turning GPU Compute Into an Asset ClassThe Verge examines Nvidia's arrangements with Goldman, BlackRock, Apollo and Blackstone that treat GPU compute as a financeable asset class, and argues the structure does not hold up under scrutiny. The concern is that depreciating hardware with a three-to-five year useful life is being underwritten like long-duration infrastructure. If you are modeling AI capex durability, the financing structure is now a bigger risk variable than chip supply.

Graph relationships

RELEASED
Salesforce
Salesforce has run the user-authored extensions model since 2007 via Apex.
Source finding
RELEASED
Fin AI
Fin AI announced Apex customer support model
Source finding
RELEASED
Fin AI
Intercom's Fin AI launched Apex custom-built model for support workflows.
Source finding
USES
AI SDK
apex is built on the AI SDK.
Source finding
BUILT BY
pensarai
apex is built by pensarai.
Source finding
BENCHMARKED AGAINST
GitHub Copilot
GitHub Copilot performance was evaluated across six APEX worlds.
Source finding
COMPETES WITH
Claude
Apex claims higher resolution rate than Claude for customer support.
Source finding
COMPETES WITH
GPT
Apex claims higher resolution rate than GPT for customer support.
Source finding
USES
Anthropic
apex uses Anthropic models.
Source finding
RELEASED
Salesforce -> apex
Salesforce has run the user-authored extensions model since 2007 via Apex.
Source finding
RELEASED
Fin AI -> apex
Fin AI announced Apex customer support model
Source finding
RELEASED
Fin AI -> apex
Intercom's Fin AI launched Apex custom-built model for support workflows.
Source finding

Source trail

arXiv 2608.18050 Mercor built this with Ramp Signed June 15 Morph LLM APEX-Agents-AA benchmark Stanford 2026 AI Index arXiv Bloomberg TechCrunch KXAN Semafor / Hacker News Dwarkesh Podcast

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…