Frontier AI Labs Keep Becoming Consulting Firms
May 12, 2026
D.A.D. today covers 10 stories from 5 sources. What's New, What's Innovative, What's Controversial, What's in the Lab, and What's in Academe.
D.A.D. Joke of the Day: My AI confidently told me the meeting was at 3pm. It was at 10am. So now I know what they mean by "artificial" intelligence.
What's New
AI developments from the last 24 hours
OpenAI Builds Its Own $4B Consulting Arm, a Week After Its PwC Deal
OpenAI wants in on the enterprise implementation business itself. The new OpenAI Deployment Company is a majority-owned subsidiary launching with $4 billion in backing — from TPG, Advent, Bain Capital, Brookfield, and 15 other investment and consulting partners — to embed engineers directly inside client organizations to identify AI opportunities, redesign workflows, and build systems. Its opening move is acquiring Tomoro, an applied-AI firm with roughly 150 'Forward Deployed Engineers,' a term borrowed from Palantir's playbook. The launch comes just a week after OpenAI announced a partnership with PwC to embed AI agents in enterprise finance teams — and mirrors Anthropic's same-day announcement on May 5, when it co-founded a competing enterprise AI services firm with Blackstone, Goldman Sachs, and Hellman & Friedman (D.A.D. covered both).
Why it matters: A week ago the segmentation looked clean — OpenAI rode shotgun with PwC into the Fortune 500 while Anthropic built its own arm for the mid-market the big firms underserve. Today's move blurs that line: OpenAI is doing both. The doubled-down bet is that shipping the model is no longer enough — and that the long-term margin lives in implementation, the messy, lucrative work of actually getting AI installed inside large organizations.
Claude Code Gets a Dashboard for Managing Multiple AI Agents
Anthropic released agent view in Claude Code, a new interface that lets developers manage multiple simultaneous AI coding sessions from a single screen. Previously, running parallel Claude Code sessions meant juggling terminal tabs and tracking mental checklists. The new view shows all active sessions at a glance—which are waiting for input, which are still running, and which are done. Users can reply to a session inline without leaving the dashboard, launch new sessions in the background, or attach directly to any session for a full transcript.
Why it matters: If your team uses Claude Code for coding tasks, this reduces the overhead of running several AI agents in parallel—a workflow that's becoming more common as teams use AI to tackle multiple workstreams simultaneously.
GitLab Cuts Jobs, Cites Shift to AI-Driven Development Era
GitLab announced a voluntary separation program as part of a broader restructuring, with the new company structure to be finalized by June 1, 2026. CEO Sid Sijbrandij framed the move as positioning for what he called the 'agentic era' of AI-driven development, describing it as the largest market shift in twenty years. The company reaffirmed its financial guidance; specifics on headcount and cost impact will come on the June 2 earnings call. GitLab said most savings will be reinvested in growth and technology.
Why it matters: Another major dev tools company is restructuring around AI—signaling that even platforms built for developers see the ground shifting beneath their business models.
Discuss on Hacker News · Source: about.gitlab.com
AI Coding Tools May Work Better With Strict Languages Like Rust
A widely-circulated essay argues that AI coding assistants have upended conventional wisdom about language choice. The traditional logic—pick Python or JavaScript for faster development despite slower performance—may no longer hold. The claim: languages like Rust and Go, with strict compilers that catch errors immediately, actually work better with AI agents because the compiler feedback helps models self-correct in real time. Supporting examples include a developer who used 16 parallel Claude agents to build a 100,000-line C compiler in Rust for about $20,000, and another who ported 25,000 lines of C++ to Rust in two weeks with zero test regressions.
Why it matters: If the thesis holds, teams may increasingly choose performance-optimized languages without the traditional productivity penalty—potentially shifting hiring needs and codebase decisions toward compiled languages that were previously considered too slow to develop in.
Discuss on Hacker News · Source: medium.com
What's Innovative
Clever new use cases for AI
Developer Built Custom Sleep Monitor in Hours With AI Coding Help
A developer built a custom sleep monitoring system over a weekend—two USB microphones, a Raspberry Pi, and a web app that correlates audio events with Garmin sleep data—to identify what sounds were waking them up. The notable part: they claim AI coding assistance compressed what would have been a "too much effort" project into roughly 8 hours of work. The author admits they didn't read the AI-generated code, treating the AI as a full implementation partner rather than an assistant.
Why it matters: This is a small but telling example of how AI coding tools are shifting the calculus on personal projects—problems that weren't worth solving become solvable when implementation time drops dramatically.
Discuss on Hacker News · Source: martin.sh
What's Controversial
Stories sparking genuine backlash, policy fights, or heated disagreement in the AI community
Sutskever Testifies Altman Showed 'Consistent Pattern of Lying'; Board Pursued Anthropic Merger After 2023 Ouster
OpenAI co-founder and former chief scientist Ilya Sutskever testified Monday that he spent roughly a year gathering evidence for OpenAI's board that CEO Sam Altman displayed a "consistent pattern of lying." Sutskever said his dossier ran to 52 pages and that Altman's conduct included "undermining and pitting executives against one another"; he discussed removing Altman with then-CTO Mira Murati for "a long time" before the November 2023 board vote. Sutskever also confirmed that after Altman's brief ouster, OpenAI's remaining board members met with rival Anthropic about a proposal for the Claude maker to merge with OpenAI and take over its leadership — a deal that never materialized. It has been previously reported that other officials tried and failed to recruit Dario Amodei to take over OpenAI. Sutskever disclosed his stake in OpenAI is worth about $7 billion today. Closing arguments in Elon Musk's lawsuit, which seeks $150 billion in damages and Altman's removal, are set for Thursday.
Why it matters: With OpenAI eyeing a potential trillion-dollar IPO, courtroom revelations about its CEO's character — and how close the company came to being absorbed by its closest rival — could reshape how investors, employees, and regulators view the lab that defined the post-ChatGPT era.
What's in the Lab
New announcements from major AI labs
ChatGPT's User Base Now Majority Women, Fastest Growth in Emerging Markets
OpenAI released Q1 2026 data showing ChatGPT's user base has broadened significantly beyond its early-adopter demographic. Users with typically feminine names now represent over half of identifiable users, and the over-35 cohort gained message share—though younger users still lead overall. Geographically, growth is strongest in emerging markets: the Dominican Republic, Haiti, Japan, Mexico, Tanzania, Brazil, and others climbed fastest in per-capita usage. Top workplace uses included content creation, health documentation, and information retrieval.
Why it matters: The data suggests generative AI is transitioning from novelty to utility—the demographic and geographic spread indicates ChatGPT is becoming infrastructure rather than experiment for a much wider slice of the workforce.
Google Finance Adds AI Research Tools, Expands Across Europe
Google is rolling out its revamped AI-powered Google Finance across Europe with full local language support. The updated platform includes AI research tools (including Deep Search for financial queries), technical charting with indicators, expanded coverage of commodities and cryptocurrencies, and live earnings call coverage with AI-generated summaries. Google says the features aim to help investors research companies and track markets without switching between tools.
Why it matters: For finance professionals and investors in Europe, this puts AI-assisted market research directly inside a free, familiar Google product—potentially reducing reliance on Bloomberg terminals or paid research platforms for routine queries.
What's in Academe
New papers on AI and its effects from researchers
Technique Helps AI Adapt Responses to Users' Cultural Backgrounds
Researchers developed DISCA, a method for making AI models respond more appropriately to users from different cultural backgrounds—without retraining the model. The technique uses simulated personas based on World Values Survey data to steer responses at runtime, requiring only API access rather than access to model internals. Testing across 20 countries and seven open-weight models showed 10-24% reduction in cultural misalignment on standardized benchmarks. The approach offers a potentially cheaper path for companies deploying AI globally without maintaining separate fine-tuned models for different regions.
Why it matters: For organizations serving international users, this suggests future AI tools may better navigate cultural differences in values and norms without expensive per-region model customization.
Best AI Agents Still Fail 40% of Real-World Tasks, Benchmark Shows
A new benchmark called WildClawBench tests AI agents on realistic, extended work tasks—averaging 8 minutes and 20+ tool calls each—using actual command-line environments rather than artificial sandboxes. The results are humbling: even the top performer, Claude Opus 4.7, solved only 62.2% of tasks, with all other frontier models below 60%. Perhaps most striking, switching which agent framework runs the same model shifted scores by up to 18 points, suggesting current AI agent performance depends heavily on the scaffolding around the model, not just the model itself.
Why it matters: For teams evaluating AI agents for complex workflows, this research suggests the tool wrapping the AI may matter as much as which model you choose—and that even leading systems fail roughly 40% of extended real-world tasks.
Method Detects When AI Vision Tools Guess Instead of Actually Looking
Researchers developed BICR, a method to detect when AI vision models are essentially guessing based on language patterns rather than actually looking at the image. The technique compares how a model responds to real images versus blacked-out images—if it's equally confident either way, it's not truly using visual information. Tested across five vision-language models on tasks including medical imaging and financial document analysis, BICR outperformed seven existing approaches at catching this problem while adding no extra processing time.
Why it matters: For organizations using AI to analyze medical scans, financial documents, or other images where accuracy is critical, this offers a way to flag when the model is confabulating from text patterns rather than genuinely interpreting what it sees.
What's Happening on Capitol Hill
Upcoming AI-related committee hearings
Wednesday, May 13 — Hearings to examine how social media verdicts demand federal action to protect kids online. Senate · Senate Judiciary Subcommittee on Privacy, Technology, and the Law (Open Hearing) 226, Dirksen Senate Office Building
What's On The Pod
Some new podcast episodes
How I AI — Spec-driven development: The AI engineering workflow at Notion | Ryan Nystrom
AI in Business — Fixing the Pilot‑to‑Production Gap in Enterprise AI - with Lawrence Whittle of HTEC Group