the Now & the Next

A bi-weekly speculative fiction suggesting the shape of things to come.
(sourced from trustworthy trade pubs, think tanks + frontier science news)

462 Signals Tracked
7 Collisions Identified
10 Frontier Science Cards
4 Signal Slices

This fortnight brought 462 signals across enterprise AI, and one pattern emerged louder than the rest: the interface is disappearing. Commerce is moving inside chatbots. Drug discovery is moving inside foundation models. Security is moving inside AI agents — and so are the attackers. Meanwhile, the humans who built all of it are being asked to take buyouts, track their keystrokes, and negotiate with their unions over who controls the bots. Here are seven collisions we detected in the noise.

01
The Chatbot Becomes the Storefront

Commerce moves inside the conversation — and the old web starts to go dark

+

Starbucks launched a beta ordering app inside ChatGPT — describe your mood, upload a photo, and the chatbot suggests your next latte. Little Caesars followed with full pizza ordering through ChatGPT across the U.S., Mexico, and Canada. David's Bridal deployed Shopify's Agentic Storefronts into both ChatGPT and Microsoft Copilot, calling it the 'Aisle to Algorithm' strategy. Ticketmaster put event discovery and ticket pricing inside ChatGPT, including sponsored ad placements. Spotify integrated personalized music and podcast recommendations into Claude. Ulta Beauty partnered with Google to make its product assortment shoppable through Gemini. Home Depot launched AI phone agents fielding store calls, resolving issues four times faster than traditional menus. Casey's expanded its SoundHound-powered voice ordering across 2,600 locations, having already processed 21 million guest interactions. And Adobe reported AI-driven traffic to U.S. retail sites grew 393% year-over-year in Q1 2026, converting 42% better than non-AI traffic. The storefront isn't dying. It's migrating into the conversation.

⚡ The Now

Brands are racing to establish presence inside AI interfaces before the discovery layer solidifies. Starbucks, Little Caesars, David's Bridal, Ticketmaster, and Spotify have all embedded commerce directly into ChatGPT, Claude, and Gemini within a single fortnight. AI-referred shoppers spend 48% more time on site and browse 13% more pages. But Adobe warns that many retail sites remain machine-unreadable — invisible to the AI agents that are fast becoming the primary shopping interface.

→ What's Next

AI storefronts become the default discovery channel within 18 months. Brands that aren't shoppable inside ChatGPT, Gemini, and Claude will become invisible to the fastest-growing consumer channel in history. SEO gives way to AEO — agent engine optimization. Retail media networks will need to buy placements inside AI conversations, not just search results. The brands that win will be the ones whose product data is machine-readable, agent-addressable, and conversationally surfaced. The rest will wonder where their traffic went.

Starbucks
Mood-based drink discovery inside ChatGPT. Upload a photo, describe your vibe, and the chatbot suggests and starts your Starbucks order.
Chain Store Age
Full pizza ordering through ChatGPT across the U.S., Mexico, and Canada. Personalized recommendations based on dietary needs, budget, and group size.
Retail Brew
Shopify Agentic Storefronts power wedding dress shopping inside AI assistants. The 'Aisle to Algorithm' strategy tracks AI channel referrals.
Spotify
Personalized playlists and podcast recommendations inside Claude. Premium users describe a mood and get a custom playlist generated on the spot.
Chain Store Age
AI-driven traffic to U.S. retail sites grew 393% YoY in Q1 2026. Converts 42% better than non-AI traffic. Many retailers still aren't machine-readable.
02
The Layoff Has a Formula

Tech companies stop pretending — the workforce math is now explicit

+

Microsoft offered its first-ever voluntary retirement buyout to roughly 8,750 U.S. employees — 7% of the domestic workforce — while simultaneously committing $80 billion to AI infrastructure. Nike cut 1,400 jobs, mostly in technology, consolidating its tech footprint to two hubs and eliminating the CTO role entirely. Meta is reportedly planning to cut 16,000 positions — 20% of its workforce — in what would be the second-largest tech layoff of 2026, while simultaneously deploying keystroke-tracking software to train AI agents on employee workflows. UBS analysts forecast over 40,000 U.S. retail stores will close in the next five years, accelerated by AI-powered e-commerce. Gartner predicts half of new warehouses built by 2030 will be designed as human-optional. And in a development unions saw coming, labor negotiations are increasingly about who controls the bots — not wages. OpenAI's own policy blueprint now advocates formal worker co-governance of AI deployment, and the No Robot Bosses Act would require human review of all automated employment decisions. The workforce isn't being disrupted. It's being redesigned around a formula.

⚡ The Now

The gap between AI investment and workforce reduction has collapsed into a single line item. Microsoft's $80B AI spend and 7% buyout are the same strategy. Meta tracks keystrokes to train the agents that will replace the employees being tracked. Nike eliminates the CTO while centralizing tech operations. The workforce math is no longer hidden behind euphemisms — it's in the earnings call. Anthropic's survey of 81,000 Claude users found that workers in AI-exposed roles report higher anxiety, with early-career workers most affected.

→ What's Next

Labor relations become the defining governance challenge of the AI era. Within 12 months, expect union contracts to include AI deployment clauses as standard. The No Robot Bosses Act or its variants will pass in at least three U.S. states. Companies that treat workforce transformation as a cost-cutting exercise will face regulatory backlash and talent flight. The winners will be the ones that build co-governance into their AI deployment — not because it's ethical, but because it's the only approach that scales without breaking.

Computerworld
First-ever voluntary retirement program. ~8,750 employees. Coincides with $80B AI infrastructure commitment. The budget tells the story.
Retail Dive
1,400 jobs cut, mostly tech. CTO role eliminated. Tech operations consolidated to two hubs under the 'Win Now' turnaround.
Computerworld
Mouse movements, clicks, keystrokes, and screen activity captured to train workplace AI agents. Part of Meta's 'Agent Transformation Accelerator.'
Material Handling & Logistics
Gartner: 50% of new warehouses in developed markets will be robot-centric with humans optional by 2030. Labor shortages accelerate the shift.
Human Resources Executive
Unions negotiating AI deployment oversight. OpenAI advocates worker co-governance. The No Robot Bosses Act would require human review of all automated employment decisions.
03
GPT-5.5 Rewires the Stack

A model drops — and every platform scrambles to govern the fallout

+

OpenAI released GPT-5.5 and GPT-5.5 Pro, its most capable models yet, across the API, Codex, and ChatGPT simultaneously. Within days, Databricks had it running inside Unity AI Gateway with full governance — centralized security, cost control, and audit trails for every GPT-5.5 call. NVIDIA deployed Codex on its GB200 NVL72 rack-scale systems, with 10,000+ employees already using it, reporting 35x lower cost per million tokens. Abridge began clinical validation of GPT-5.5 for healthcare documentation, testing it across note generation and clinical decision support. DeepSeek dropped V4 with a million-token context window — built specifically for agentic workloads that need to hold an entire codebase in memory. Codex crossed 4 million weekly active developers and launched enterprise scaling through partnerships with Accenture, Capgemini, and Infosys. And Databricks released Agent Bricks, a governed enterprise agent platform adopted by thousands of organizations including Workday, Virgin Atlantic, and AstraZeneca. The model is only half the story. The governance layer is where the real race is happening.

⚡ The Now

GPT-5.5 triggered the fastest enterprise integration cycle in AI history. From launch to governed enterprise deployment in under a week. Databricks wrapped it in audit trails and cost controls before most companies finished reading the press release. NVIDIA's internal deployment proved the economics — 35x cheaper per token, 50x more efficient per megawatt. Meanwhile, Databricks' Agent Bricks and Unity AI Gateway signal that the governance stack is hardening around agent orchestration, not just model access.

→ What's Next

The 'model release + governance scramble' cycle becomes permanent. Every frontier model drop will trigger a 72-hour race among platform vendors to wrap it in enterprise guardrails. Companies that don't have a governance layer ready — cost controls, audit trails, identity-based access, fallback routing — will find themselves unable to safely adopt new models at speed. The competitive advantage shifts from 'who has the best model' to 'who can govern it fastest.' Expect agent governance to become the most contested product category in enterprise software by year's end.

OpenAI
GPT-5.5 and GPT-5.5 Pro launch across API, Codex, and ChatGPT. Enhanced agentic capabilities, stronger safety, and significantly improved coding and knowledge work.
Databricks
GPT-5.5 available inside Unity AI Gateway with centralized security, cost control, audit trails, and failover. Enterprise governance from day one.
NVIDIA Blog
10,000+ NVIDIA employees use Codex on GB200 NVL72 systems. 35x lower cost per million tokens. 50x higher throughput per megawatt.
Hugging Face
Million-token context window built for agentic workloads. KV cache reduced to 7-10% of previous versions. The open-source answer to Codex's enterprise push.
Databricks
Enterprise agent platform with multi-model support, identity-first security, and governance. Adopted by Workday, Virgin Atlantic, AstraZeneca, and thousands more.
04
Pharma Bets the Lab on AI

Billion-dollar partnerships signal that drug discovery's operating system is changing

+

Merck struck a partnership with Google Cloud worth up to $1 billion, deploying agentic AI across R&D, manufacturing, commercial, and corporate functions — the largest enterprise AI deal in pharmaceutical history. Novo Nordisk partnered with OpenAI to embed AI across drug discovery, manufacturing, supply chain, and corporate operations, with strict data protection and human oversight requirements. OpenAI went further, launching GPT-Rosalind, a biotech-specific reasoning model designed to accelerate biology research and translational medicine, built in collaboration with Amgen, Moderna, and other biopharma partners. Eli Lilly announced a $7 billion acquisition of CAR-T developer Kelonia while simultaneously entering a $2.75 billion collaboration with Insilico Medicine for AI-originated therapeutics. Amazon fine-tuned its Nova models with Nimbus Therapeutics for molecular-property prediction, creating a conversational AI that can predict drug properties and explain its reasoning. And researchers at Houston Methodist developed CAMPER, an AI platform that designed a peptide effective against MRSA — antibiotic-resistant bacteria — in laboratory tests. The pharmaceutical industry isn't adopting AI as a tool. It's adopting AI as an operating system.

⚡ The Now

The pharma-AI partnership model has leapt from pilot to platform in a single quarter. Merck's $1B Google deal and Novo's OpenAI partnership represent the largest enterprise AI commitments in pharmaceutical history. These aren't point solutions — they're full-stack transformations spanning R&D, manufacturing, and supply chain. Meanwhile, GPT-Rosalind marks the first time a frontier AI lab has built a purpose-specific model for drug discovery. The typical 10-15 year drug development cycle is the target.

→ What's Next

AI-native drug discovery pipelines will produce the first wave of clinically validated compounds within 24 months. The convergence of billion-dollar pharma partnerships, purpose-built biotech models, and AI-designed therapeutics like CAMPER's anti-MRSA peptide means the proof-of-concept phase is ending. Companies without an AI drug discovery strategy will find themselves competing against pipelines that move at computational speed. The pharmaceutical industry's cost structure — and its competitive dynamics — will be permanently altered.

Fierce Pharma
Up to $1 billion for Google Cloud's agentic AI platform across R&D, manufacturing, and corporate. Google engineers embedded in Merck teams.
Fierce Pharma
Enterprise-wide OpenAI deployment from drug discovery to supply chain. Strict data protection and human oversight built in from the start.
Fierce Biotech
First biotech-specific reasoning model. Built with Novo, Amgen, Moderna. Targets the 10-15 year drug development timeline.
Drug Discovery & Development
$7B for CAR-T capability plus a $2.75B collaboration with Insilico Medicine for AI-originated therapeutics. Lilly bets on both biology and computation.
EurekAlert!
CAMPER platform designed an antimicrobial peptide effective against MRSA in lab tests. AI goes from predicting drug targets to designing the drug.
05
The Humanoid Punches In

Bipedal robots get work addresses, shift schedules, and SAP integrations

+

Siemens and startup Humanoid, using NVIDIA's physical AI stack, deployed the HMND 01 Alpha humanoid robot in Siemens' electronics factory in Erlangen, Germany — autonomously handling tote logistics and meeting all performance benchmarks. The same week, Accenture, SAP, and Vodafone ran a humanoid robot pilot in a German warehouse, with robots performing autonomous visual inspections and feeding findings directly into SAP's warehouse management system in real time. Siemens separately launched its Eigen Engineering Agent, an AI that autonomously plans, executes, and validates industrial automation engineering tasks — delivering 2-5x faster execution across pilot deployments in 19 countries. Google and Boston Dynamics upgraded Spot with DeepMind's Gemini Robotics-ER 1.6, giving the quadruped embodied reasoning for autonomous hazard detection and instrument reading. In agriculture, Bayer deployed robotic security dogs across 8,000 acres in Hawaii and California. And Agtonomy and Kubota are embedding physical AI into existing tractors to address farm labor shortages — not replacing farmers, but enabling one operator to manage multiple autonomous machines. The humanoid isn't a prototype anymore. It has a badge and a shift.

⚡ The Now

Physical AI has moved from demonstration to deployment in factories, warehouses, fields, and facilities. Siemens' factory deployment met all production metrics. Accenture's warehouse pilot integrated directly with enterprise systems. Boston Dynamics' Spot now reasons about its environment autonomously. Each deployment pairs AI perception with physical actuation in environments that were previously automation-resistant. The common thread: these robots don't replace workers — they fill roles that can't be staffed.

→ What's Next

Humanoid robots become a procurement category within 24 months. As Siemens, NVIDIA, and Google converge on the physical AI stack, the humanoid shifts from a robotics curiosity to an enterprise line item. Companies will evaluate humanoids the way they evaluate enterprise software — by TCO, integration complexity, and SAP compatibility. The labor conversation moves from 'will robots take jobs?' to 'which shifts can't we fill without them?' The factory of the near future isn't lights-out. It's mixed-shift.

Siemens
HMND 01 Alpha humanoid performing autonomous tote logistics in Siemens' Erlangen factory. NVIDIA's Jetson Thor, Isaac Sim, and Isaac Lab power the stack.
Assembly
Accenture-SAP-Vodafone warehouse pilot. Humanoids perform inspections and feed findings into enterprise systems in real time.
Siemens
Autonomous industrial automation engineering. 2-5x faster execution. Piloted with 100+ companies across 19 countries.
IEEE Spectrum
Spot upgraded with Gemini Robotics-ER 1.6 for embodied reasoning. Autonomous hazard detection, instrument reading, and environment interpretation.
AgFunderNews
Agtonomy and Kubota embed physical AI in existing tractors. One operator manages multiple autonomous machines. Augmentation, not replacement.
06
The Agent Attack Surface

AI agents create the biggest new vulnerability class since the cloud

+

A critical authentication flaw in Microsoft's Azure SRE Agent allowed unauthorized accounts to silently eavesdrop on live command streams, internal reasoning, and deployment credentials across tenants — the first major vulnerability in an enterprise AI agent system. IBM responded by launching Autonomous Security, a multi-agent AI protection system that detects and mitigates AI-driven threats at machine speed, alongside cybersecurity assessments specifically designed for frontier model threats. Anthropic launched Project Glasswing, partnering with AWS, Google, Microsoft, NVIDIA, and Cisco to embed continuous AI-driven security scanning across their platforms — a shift from periodic audits to self-healing infrastructure. Mozilla deployed Claude Mythos Preview to find vulnerabilities in Firefox, fixing 271 latent security issues in a single release. OpenAI launched Trusted Access for Cyber, a $10 million program giving Bank of America, JPMorgan, CrowdStrike, and Cisco access to GPT-5.4-Cyber for defensive operations. Siemens launched managed detection and response for energy infrastructure. And the banks themselves reported that 80% of executives now include cybersecurity in their AI budgets. The agents are powerful. The agents are also vulnerable. Both things are true at the same time.

⚡ The Now

AI agents have become both the most powerful defensive tool and the newest attack surface in enterprise security. Azure's SRE Agent flaw exposed credentials across tenants. IBM launched agent-specific threat detection. Anthropic's Glasswing coordinates vulnerability discovery across the entire tech ecosystem. Mozilla proved that AI can find hundreds of bugs humans missed. The defensive potential is real — but so is the risk. Every enterprise deploying AI agents is simultaneously deploying a new class of vulnerability.

→ What's Next

Agent security becomes a board-level concern within 12 months. As AI agents gain write access to production systems, identity management, and sensitive data, the blast radius of agent vulnerabilities will dwarf traditional software bugs. Expect a new category of security products — agent firewalls, agent identity management, agent behavioral monitoring — to emerge as a billion-dollar market. The companies that treat agent security as an afterthought will learn the lesson the way the industry always learns security lessons: the hard way.

Network World
Critical multi-tenant auth flaw exposed live command streams, internal reasoning, and credentials. The first major enterprise agent vulnerability.
Network World
IBM Autonomous Security: multi-agent AI system for real-time detection and remediation of AI-driven threats. Plus frontier model threat assessments.
Data Center Knowledge
Claude Mythos model deployed to find vulnerabilities across AWS, Google, Microsoft, NVIDIA, and Cisco platforms. Continuous scanning, not periodic audits.
Mozilla
Firefox 150 includes fixes for 271 vulnerabilities discovered by Claude Mythos Preview. AI-driven defense outpaces manual and dynamic analysis methods.
Banking Dive
80% of banking executives include cybersecurity in AI budgets. JPMorgan and Morgan Stanley testing Mythos Preview for defense. AI spending up 33% QoQ.
07
The Autonomous Warfighter

Military autonomy crosses the line from experimental to operational

+

The U.S. Army's CH-47F Chinook completed its first fully automated landing without a pilot onboard — using Boeing's Approach-to-X software while the Army simultaneously cuts 6,000 aviation personnel. Northrop Grumman completed eight flight tests of its Talon IQ platform, proving the ability to hot-swap AI control systems mid-flight — switching between autonomy software from Shield AI, Accelint, and Applied Intuition without disrupting flight controls. Lockheed Martin invested $25 million in Fortem Technologies for AI-driven counter-UAS systems integrated into its Sanctum ecosystem, reducing drone engagement costs by over 80% compared to kinetic interceptors. France modified the Rafale's fire control system to let its 30mm cannon engage Iranian-style Shahed drones — adapting a $100 million fighter jet to counter $50,000 drones. And NATO selected Saab's GlobalEye over Boeing's E-7A Wedgetail for its next-generation AWACS fleet, a €5+ billion shift toward European defense industrial autonomy. The common thread: military systems are being redesigned around autonomous operation, not as a feature but as a structural requirement driven by personnel cuts, asymmetric threats, and the speed of modern conflict.

⚡ The Now

Military autonomy has moved from R&D to operational deployment across air, land, and counter-drone missions. Chinooks land themselves. Fighter jets swap AI brains mid-flight. Counter-drone systems engage at 80% lower cost. Each development responds to the same pressure: modern warfare moves faster than human decision cycles, with cheaper weapons than human-piloted platforms can justify. The U.S. Army is cutting pilots while investing in autonomous landing. That's not coincidence — it's doctrine.

→ What's Next

Autonomous military systems become the primary force structure within five years. The interoperability demonstrated by Talon IQ — plug-and-play AI modules across platforms — will become the standard architecture for allied autonomous systems. NATO's shift to GlobalEye signals European defense industrial independence accelerating. Counter-UAS economics will reshape procurement budgets as militaries realize they need cheap autonomous defenders, not expensive piloted interceptors. The next conflict between peer adversaries will be the first where autonomous systems outnumber human operators on both sides.

Task and Purpose
CH-47F completes fully automated approach and landing. Boeing's A2X software. Concurrent with 6,000 Army aviation personnel cuts.
Breaking Defense
Eight flight tests proving mid-flight AI control handoffs. Modular open architecture for plug-and-play autonomy software across platforms.
Lockheed Martin
AI-driven counter-UAS in the Sanctum ecosystem. 80%+ cost reduction over kinetic interceptors. Open architecture for allied interoperability.
Janes
Rafale fire control updated to engage Shahed-type drones with its 30mm cannon. A $100M jet adapted to counter $50K weapons.
Defense News
Saab GlobalEye selected over Boeing for €5B+ AWACS replacement. European defense autonomy accelerates as U.S. programs face delays.

Frontier Science Feeding the Machine

The research signals underneath the enterprise stories. These breakthroughs — in chip design, brain mapping, thermoelectrics, clinical AI, and environmental prediction — are the tectonic plates on which the business collisions above are riding.

Autonomous Engineering
Verkor.io's Design Conductor autonomously designed a working RISC-V CPU core in 12 hours. LLMs guided through structured human review loops, debugging RTL, running simulations, and iterating. Chip design joins the list of engineering disciplines where AI sets the clock.
Neuroscience
Google DeepMind generates synthetic neuron images to train AI models that trace neural connections in 3D brain scans. Connectomics — the mapping of every synapse — accelerates from years to months.
Energy Harvesting
Japanese researchers built an AI tool that designs thermoelectric generators 10,000x faster than traditional methods. Waste heat from factories, engines, and data centers becomes a viable power source.
Clinical AI
Stanford Medicine study shows LLM chatbots improve physician decision-making on complex clinical management questions beyond diagnosis. The AI doesn't replace the doctor — it improves the doctor.
Cancer Detection
City of Hope and UC Berkeley developed a microfluidic platform that mechanically stresses individual breast cells, then uses AI to assess cancer risk at the single-cell level. Detection moves from tissue to cell.
Neuroprosthetics
USC, UCI, and Caltech built a fully implantable brain-computer interface that lets patients control robotic exoskeleton legs with thoughts — and receive sensory feedback. The loop closes.
Seismology
ML algorithms identified over 60,000 seismic events during the 2025 Santorini volcanic sequence — far more than conventional methods detected. The earth's signals, finally readable at scale.
Environmental AI
AI is transforming environmental science from observational to predictive and precision-driven. Ecological forecasting, pollution modeling, and climate response — all moving from rear-view to windshield.
Bioacoustics
UNSW researchers built a deep learning model that detects blue whale songs in decades of ocean recordings using only a single example call. One song trains the system. Decades of hidden data surface.
Rail Safety
Chinese researchers monitor railway safety by analyzing vibrations in existing underground fiber optic cables. No new sensors needed — the infrastructure is already listening. It just needed AI ears.