Search the archive

See how it works → lightfield.app

THE APP LAYER IS DEAD. LONG LIVE THE APP LAYER

If AI made your main feature free, what would be left of your company? Apps that just wrap a model are easy to replace. The ones that last own a number their customer cares about and keep making it better. Like spreadsheets, cheaper AI will create more work, not less. Labs will build a generic support agent that handles 30 to 40 percent of tickets, but you win above that by knowing the messy details labs won't bother learning. Meta is proof that advertisers pay for sales, not for the dashboard.

THE CARDINAL SIN OF PLATFORM BUILDING; THE FDE WARS INTENSIFY

The platforms that win the next era will look far more like hyperscalers than the SaaS platforms of the past. Companies that centralize important data are in the running to be a platform in the AI era. The last generation of winners are committing the cardinal sin of starting with the platform before they've built the killer app. Doing this almost guarantees they will get the abstractions and interfaces wrong because they're optimizing them against a usage pattern that doesn't exist yet.

THE DISTRIBUTION ERA

The best B2B software companies of the next decade will build distribution from day 0. AI's development capabilities mean that the durable advantage has moved to audience and the distribution built to reach that audience. Markets are being won earlier and more convincingly than ever. A business' go-to-market is what builds their moat and dictates their growth trajectory.

MORE TOKENS IS NOT A BUSINESS OUTCOME

Counting tokens measures effort, not results, just like counting hours worked. Companies are done experimenting with AI for free and now want proof it pays off. The first phase was cheap because labs ate the cost. Now, they charge real money, and CFOs are asking hard questions at renewal. Smart buyers use cheaper models for easy work and save the expensive ones for tasks tied to revenue. AI usage will keep growing fast, so the real question is who actually makes money from it.

HOW TO HIT $10M ARR IN A FEW MONTHS

Your best channel might already be getting worse. Treat marketing like investing. Every channel works for a while, then slowly stops paying off as you spend more. The real question isn't, "Is this channel working?" It's, "Is the next dollar still worth spending here?" The biggest wins are usually small and don't scale, like one newsletter sponsorship that converts at 8x. An "influencer marketing" hire is really three jobs in one: outreach, creative, and budget decisions.

HEADROOM

Headroom compresses everything an agent reads before it reaches the model so that users receive the same answers at a fraction of the tokens. It is suitable for running AI coding agents and working across multiple agents that have shared memory. Reverse compression is always achievable via CCR. Headroom doesn't work in sandboxed environments where local processes can't run.

DELIBERATE

Deliberate records every option an agent considers, rejects, and executes and provides an audit trail that users can export. It is built for teams running LangGraph or OpenAI Agents pipelines in production. Deliberate wraps agent loops and writes a complete record that compliance teams can sign off on.

THE MARKET FOR LONG HORIZON TASKS

There's a strange wedge hiding here, building the hard tasks labs use to test their agents. A convincing fake of SAP, realistic enough to evaluate an agent inside, was rumored at $500,000, and METR says it can barely find long-horizon tasks good enough to use, with decent ones at $20,000 each. Cheap labeling is everywhere, but captured expert work that's realistic enough to fool an agent and clean enough to grade is scarce, so the eval bottleneck is domain realism, not model access.

THE GTM STACK ANTHROPIC USES FROM ITS HEAD OF INDUSTRIES

Eleanor Dorfman, Anthropic's Head of Industries, recently walked through the company's GTM stack at SaaStr AI Annual 2026. The company uses the same six tools that many other leading B2B + AI companies use, as well as Jira, Intercom Fin, Snowflake, BigQuery, and G Suite. This article details what the company does with each part of its stack.

pay attention. Learn more about sponsorship opportunities.

YOUR CUSTOMERS' AI AGENTS ARE COMING FOR YOUR DATA. ARE YOU READY?

AI will change the relationships customers will have with financial institutions, whether they decide to build anything with AI or not.

ADAPTING TO AI: WRITE THINGS DOWN

Organizations that have a culture of real writing will outperform those that do not.

CODE ISN'T PRODUCT

AI is scaling engineers' ability to create products and features, but it isn't scaling customer attention.

TEAM TOPOLOGIES AS THE INFRASTRUCTURE FOR AGENCY

Claim your savings

APPLE TO OVERHAUL IOS 27 SIRI, AI FEATURES: HERE'S A FIRST PEEK

Apple plans to announce a revamped Siri interface, a new chatbot-style app, and other major iOS 27 changes at its Worldwide Developers Conference on June 8. The launch will be a pivotal moment for the company's AI strategy. Apple previewed some of the features in 2024, but it suffered a series of delays that put it further behind competitors and hurt its reputation. The Siri revamp will likely be CEO Tim Cook's final major product launch before handing leadership over to John Ternus.

CLAUDE OPUS 4.8: “A MODEST BUT TANGIBLE IMPROVEMENT”

Claude Opus 4.8's most prominent improvement is its ability to be honest. It is more likely than previous models to flag uncertainties about its work and less likely to make unsupported claims. The model has a new feature that allows mid-conversation system messages. This means users can append updated instructions later in long-running conversations without restating the full system prompt.

THE MOST SPECTACULAR ROCKET EXPLOSION SINCE N1 JUST HAPPENED IN FLORIDA

A spectacular explosion occurred during a static fire test of Blue Origin's New Glenn rocket last night. The super heavy lift rocket exploded shortly after engine ignition. No one was injured during the failure, and the cause is still being investigated. A video of the explosion is available in the article.

WAYMO OPENS OJAI ROBOTAXIS TO SELECT RIDERS AS COMPANY AIMS TO LOWER COST OF FLEET EXPANSION

Waymo's new Ojai robotaxis are designed to work better in snowier conditions. They are cheaper to manufacture and feature the sixth-generation Waymo Driver, which can better detect critical details in low-light situations. The vehicles will be available to some public passengers in San Francisco, Los Angeles, and Phoenix, with service expanding to San Diego, Las Vegas, and Denver in the summer. Waymo plans to have thousands of Ojai cars on the road by the end of the year.

Vercel, and Clay use Flow daily. Try it free

INTRODUCING DYNAMIC WORKFLOWS IN CLAUDE CODE

Dynamic workflows in Claude Code enable Claude to take on challenging tasks end-to-end. Claude dynamically writes orchestration scripts that run hundreds of parallel subagents, checking the work before returning it to users. The feature is available now in research preview in the Claude Code CLI, Desktop, and the VS Code extension for Max, team, and Enterprise plans. It can consume substantially more tokens than a typical Claude Code session, so it is recommended that users start with a scoped task to get a feel for usage.

THE ORCHESTRATION TAX

Spawning agents is easy. The real skill is in designing the system around human attention. Agents can make you feel productive while barely producing anything. This failure mode will be invisible until the system breaks and the developer realizes they have no idea how it works anymore.

HOW META IS LOOKING FOR REVENUE OUTSIDE ADVERTISING TO JUSTIFY ITS BALLOONING CAPEX BILL

Meta's capital expenditures are forcing it to explore new revenue sources. The company has announced a series of subscription tiers and it is monetizing its AI chatbot. It is also launching a new Enterprise Solutions unit, which will mimic the forward-deployed engineer models used by Google and OpenAI to open up B2B revenue streams. The company is considering opening its own cloud computing business once it has excess compute.

HOW ONLINE SLEUTHING HELPED CATCH THE ‘GOOGLE INSIDER' ON POLYMARKET

Polymarket doesn't require users to submit proof of identity, making it mostly anonymous. However, it uses blockchain-based technology, which makes it possible for anyone to monitor the bets, profits, and losses of individual accounts. This allows sleuths to spot suspicious activity and call it out, even if they can't identify the actual traders. Online sleuths have become an unexpected ally of law enforcement in tackling the emerging challenge of insider trading on prediction markets. Blockchain trading is transparent and traceable - even sophisticated actors have trouble staying hidden.

WE'RE IN THE OVER-ENGINEERING GAME NOW

AI allows developers to over-engineer everything as the vision for the application is no longer constrained by resources.

THE RADICAL NETWORK REDESIGN THAT LED AWS TO FORGE A MORE RESILIENT CLOUD

A look at how a team at Amazon applied random graph theory to a data center network.

IBM RALLIES AFTER COMMITTING TO SPEND $10 BILLION TO PURSUE THE HOLY GRAIL OF QUANTUM COMPUTING

IBM plans to spend over $10 billion over the next five years to create a large-scale, fault-tolerant quantum computer.

BE TLDR'S FIRST VP MARKETING ($300K-$400K, FULLY REMOTE)

TLDR is hiring a VP of Marketing to scale growth, pipeline, and brand across 8M+ subscribers, 13 newsletters, and advertisers like AWS, Google Cloud, Microsoft, and Anthropic. Apply here [24].

ANTHROPIC NEARS $1 TRILLION VALUATION IN LATEST ROUND

Anthropic is now the most valuable AI company in Silicon Valley.

PACKAGE MANAGERS THAT PACKAGE PACKAGE MANAGERS

A look at 42 package managers and whether they install other package managers.

JD.COM FOUNDER VOWS TO PROTECT CHINESE JOBS FROM AI AND ROBOTS

The company has promised to prevent its 900,000-strong workforce from losing their jobs to automation.

Try Wispr Flow Free [6] | Get Flow

ANTHROPIC RAISED $65B IN SERIES H FUNDING

Anthropic announced a $65 billion Series H round at a $965 billion post-money valuation, citing strong enterprise adoption, $47 billion in run-rate revenue, and plans to expand compute capacity, research, and product development.

OPUS 4.8

Anthropic released Claude Opus 4.8 with benchmark improvements, adjustable effort controls, dynamic workflows in Claude Code, and a faster mode that became significantly cheaper.

HOW LONG IS ANTHROPIC'S LEASE WITH SPACEX? OPINIONS VARY

SpaceX earlier this month signed a major compute deal with Anthropic worth billions of dollars a month. However, Elon Musk recently downplayed the deal, saying that SpaceX had not committed to leasing its compute for years, even though it is possible that might happen. The agreement is actually a 180-day lease with a 90-day mutual cancellation thereafter. The short-term agreement was SpaceX's request as it may want the compute back at some point. Musk's statement directly contradicts SpaceX's S-1 filing, which presents the deal as a three-year agreement.

MICROSOFT TRIES TO GET BACK IN THE AI CODING GAME WITH NEW MODEL

Microsoft is developing a new AI model to strengthen its position in the AI coding arena. This effort highlights Microsoft's ongoing competition in AI development and its response to evolving industry demands. The initiative aims to enhance coding capabilities and support advancements in AI technology.

AGENT JUDGE: SOLVING LONG-CONTEXT EVALS FOR PRODUCTION AGENTS

Agent Judge improves evaluations for long-context, production agents by focusing on Search, Verification, and Adaptation. It tackles LLM judges' shortcomings by navigating long trajectories, verifying stateful actions against systems, and updating rubrics based on real feedback. Test results show Agent Judge, especially with refined rubrics, surpasses traditional LLM judges in accuracy and consistency, particularly in challenging scenarios.

HOW FAR BEHIND ARE OPEN MODELS?

Open models are generally not as capable as the best closed models. However, they aren't too far behind, with tests showing that they are only four to six months behind on public benchmarks. The gap was the smallest around the time of DeepSeek R1. It has since been growing.

INTRODUCING DYNAMIC WORKFLOWS

Jarred Sumner used dynamic workflows to rewrite Bun from Zig to Rust, achieving 99.8% test suite success with 750,000 lines of Rust in 11 days. Dynamic workflows involve Claude breaking tasks into subtasks, with agents running in parallel until results converge.

THE CURSOR DEVELOPER HABITS REPORT

Models now utilize more context to understand codebases, which reduces costs as input and cache-read tokens are cheaper than output tokens. This context-driven approach improves code calibration, increasing developer productivity and diff survival rates.

FOR OVER A DECADE, WE'VE ACCEPTED THAT END-TO-END BACKPROP IS THE ONLY WAY TO TRAIN DEEP NETWORKS

Holding the entire network in memory at once is why AI training is hitting a resource wall. Sakana Labs has found a new way to break the network into blocks and train them independently. The trick was to treat the network's forward pass like a diffusion model denoising a signal. This slashes the memory needed to train deep models.

MULTI-AGENT WORLD MODELS

NVIDIA γ-World is a generative world model that supports independently controllable, permutation-symmetric agents and delivers real-time rollouts with zero-shot generalization from two-player to four-player settings.