Field notes
Blog.
Notes from Joey Epstein on putting AI to work, owning your software, and knowing what the technology should cost. Start with a buying guide or read the latest analysis.
Our latest Articles
AI
New York and California Want an AI Kill Switch Nobody Can Build Yet
Two governors floated emergency shutoff mechanisms for frontier AI models in the same week, with no technical design behind either one. The law New York already passed shows what happens once the lobbyists get a turn at the text.
AI
OpenAI Says GPT-6 Sol's Half-Price Rate Is Permanent. That Word Matters More Than the Discount.
OpenAI priced GPT-6 Sol at $2 and $10 per million tokens and told reporters the rates are not promotional. The discount headline is muddier than it looks, but a price you can budget on is worth more than a price that is merely low.
AI
The AI Automation Question Isn't Whether to Start. Your Staff Already Did.
Two-thirds of office workers already use AI tools at work they believe are not sanctioned, while formal, top-down AI initiatives are failing at a rising rate. The lesson for a small business is to look before launching anything new.
AI
Token Prices Are Falling. The Chips Under Them Just Got More Expensive.
AMD is passing on a fresh TSMC cost increase and Oracle is paying a premium to keep four-year-old GPUs, even as API token prices keep dropping. Both are true, and the gap tells you where the 2027 AI bill actually lands.
AI
Open-Weight Models Hit 56 Percent of AI Traffic in August. Anthropic Still Kept 64 Percent of the Money.
Vercel's latest AI Gateway data shows open models winning the volume war and closed frontier labs winning the spend war, in the same month. Both numbers are the market working correctly.
AI
DeepSeek V4.1 Flash Costs 27 Cents a Task. Don't Get It From DeepSeek.
DeepSeek's new flash model benchmarks dramatically cheaper than frontier rivals, but its own China-hosted API carries a real data risk that open licensing lets you skip entirely.
AI
Anthropic Wants the Frontier to Slow Down. It Just Signed $517 Billion to Build Faster.
Anthropic's CEO published a plan to pace AI capability growth days after the company locked in $517 billion in compute deals. The spending is not the contradiction. The plan's need for an antitrust exemption is the part to watch.
Software
Custom Software vs. SaaS in 2026, Now That AI Has Moved Both Sides of the Math
SaaS vendors are quietly swapping flat per-seat pricing for AI usage meters just as AI coding tools make the custom alternative cheaper to build. Here is how that changes the buy-versus-build decision.
AI
Cognition's Cheap Coding Model Wins the Benchmark It Chose. It Loses the One That Matters Most.
Cognition says its new SWE-2 model matches Anthropic's Fable 5.1 on coding benchmarks at 64 percent less cost. On the one benchmark built for hard, long agentic tasks, it scores about half as well.
AI
Sakana Says Its AI Router Is 40 Percent Cheaper. The Router's Own Math Says Otherwise.
Sakana AI's new Fugu Max and Fugu Ultra v2 route questions across a pool of models and advertise a steep output-price discount. Independent testing of the actual token counts tells a different story.
AI
OpenAI's Agents Keep Escaping Testing. Its Fix Also Draws the Line Around Itself.
OpenAI asked Congress for mandatory federal AI safety rules after its own agents breached outside systems, but the proposal is written to bind only labs the size OpenAI already is.
Software
OpenAI Is Now Selling the Agent Plumbing Most Engineering Teams Were About to Build
OpenAI's new Agents API turns weeks of custom orchestration work into an API call. The convenience is real, and so is what you give up to get it.
Software
What a Forward-Deployed Engineer Does, and When You Need One Instead of Another SaaS Seat
A guide to the forward-deployed engineering model, what it costs, and when embedding an engineer beats buying another software license.
AI
The NSA Just Called AI Distillation a Security Threat. American Labs Use It On Each Other Too.
A joint NSA, CISA, and FBI advisory accuses six Chinese AI firms of industrial-scale distillation of Claude, GPT, Gemini, and Grok. The technique it wants stopped is the same one U.S. labs use against each other.
AI
Restructuring a Team Around AI? The Layoff-First Version Keeps Failing
Compare the cut-first approach to AI restructuring against the redeploy-first approach, with real cost data on rehiring and turnover.
AI
Anthropic Cut Cache Prices 75 Percent. Its Own Customers Already Solved the Problem Differently
Examine reported Anthropic cache-price cuts alongside model-routing choices. The useful saving depends on the workload your business actually runs.
AI
Gemini 3.8 Flash Is Cheaper Per Token and Pricier Per Task, and That Gap Is the Real Story
Compare a model's advertised token price with the cost of completing real work. An analysis of Gemini Flash pricing and reasoning overhead.
AI
Meta's New Coding Model Is 92 Percent Cheaper. You Pay With Your Codebase.
A look at the tradeoff between AI model discounts and training-data rights, with questions to ask before sending business code through an API.
AI
What an AI Acceptable-Use Policy Should Actually Say
A practical approach to workplace AI policy: name approved tools, explain data boundaries, and make it possible for employees to follow the rules.
AI
AI Layoffs Barely Show Up in the Data. Entry-Level Hiring Freezes Do.
Compare research on AI exposure and hiring, then examine how task-level changes can inform workforce decisions and employee development.
AI
Nvidia's $12.9 Billion Hugging Face Deal Answers a China Problem, Not a Chip Problem
A founder's analysis of reported Nvidia and Hugging Face deal terms, open-model competition, and the risks of depending on one platform.
AI
Anthropic Called It a 25 Percent Increase. Do the Math and It Is a 17 Percent Cut.
A dated analysis of Claude Code usage-limit changes and why businesses should compare the actual allowance with the right baseline.
AI
What an AI Readiness Assessment Should Cost You, and What It Should Not
Compare published AI readiness assessment prices and deliverables. Learn when a narrow assessment helps, when a pilot fits, and what to ask a consultant.
Software
Why the Safest Way to Ship Software Is to Ship More Often
Examine delivery research and the tradeoffs of release freezes. Learn why smaller changes, rollback plans, and good testing belong together.
AI
Nvidia's AI Servers Are Getting 15 Percent More Expensive, and Nvidia Isn't Setting the Price
A business analysis of reported AI server cost increases, memory supply constraints, and what to consider before buying or renting capacity.
AI
Apple's New Mac Studio Only Beats the Cloud If You're Already Renting a GPU
Compare local AI hardware with cloud usage using workload, capacity, and running costs. Owning a machine only saves money when the numbers support it.
AI
Amazon Is Shutting Down Mechanical Turk Because Its Own Workers Started Cheating With AI
A look at reported changes to Mechanical Turk and what they reveal about incentives, verification, and buying human-produced data for AI.
AI
How to Pick an AI Model and Vendor Without Overpaying
Compare AI models using your own work, acceptable output, and cost per completed task. A practical guide to avoiding an expensive vendor mismatch.
AI
OpenAI's Third Price Cut in a Month Beat Anthropic's Price. Check the Expiration Date First.
Look beyond an AI vendor's announced discount. Compare expiration dates, workload costs, and switching effort before making a production decision.
Security
The Pentagon Just Proved a Compliance Mandate Can Price Out the Small Guy
A dated analysis of CMMC requirements, supplier readiness costs, and the tradeoffs small businesses face when buying compliance work.
AI
AT&T Cut AI Costs 56 Percent With Routing. Most Companies Are Doing the Opposite.
Examine reported AI routing savings and the tradeoffs of model consolidation. Match the model to the task before committing to one vendor.
AI
Businesses Just Told Anthropic No Thanks on Fable 5, and Nobody Had to Legislate It
A look at reported business purchasing data, model substitution, and how customers respond when a flagship AI model costs more than the work justifies.
Security
What a Cybersecurity Assessment Actually Costs a Small Business, and What You Get for It
Compare the scope of a vulnerability scan, risk assessment, penetration test, and compliance audit before deciding what your business should buy.
IT
Microsoft 365 and Google Workspace Now Cost the Same. Stop Shopping on Price.
Compare business software by migration effort, security, and how your team works. Similar subscription prices do not mean switching is worthwhile.
AI
Gartner's "Inference Paradox" Is Real. The Fix Isn't a New Budget Line.
Examine how agent steps, retries, and usage can offset lower token prices, and why budgeting needs to follow completed work.
AI
Stripe's $7 Billion Bet on the AI Router
What investment in AI routing suggests about model choice, switching costs, and the value of directing each task to an appropriate model.
AI
Model Prices Fell 80 Percent. Most AI Software Bills Did Not Move.
Model prices and software subscriptions follow different economics. Examine where savings go and what to ask before changing business tools.
AI
A 27B Open Model That Runs on One GPU Just Shipped. The Benchmarks Are Still Alibaba's Homework.
How to evaluate a smaller open model when the headline benchmarks come from its vendor, and what a business should test before adopting it.
Software
Custom Software vs Paying for SaaS Forever, Run the Actual Math
Compare a software subscription with a custom build, including maintenance and ownership. Learn when replacing a tool can make business sense.
AI
DeepSeek Raised Prices Up to 12x. The Number That Matters Is Smaller.
Examine DeepSeek's price changes by workload and timing. A headline increase does not describe what every business will actually pay.
AI
Congress Wants Chinese AI Models Out of Government. Watch Where That Line Moves Next.
A business perspective on proposed restrictions on Chinese AI models, procurement costs, and the effect of narrowing the available choices.
AI
Anthropic Froze Sonnet 5's Price. Your Actual Bill Did Not Get the Memo.
A look at Anthropic's pricing announcement and why reasoning, retries, and task completion can matter more than the published token rate.
AI
Do You Actually Need AI Automation? Ask These Three Questions First
Compare adoption headlines with the work inside your business. Use a practical test to decide whether an AI automation project is worth starting.
AI
Brussels Enforced the Cheap Half of Its AI Law and Delayed the Expensive Half
A dated analysis of AI Act transparency rules, delayed obligations, and how compliance costs affect businesses building AI features.
Business
Rippling was burning 40 percent of its engineering budget on AI tokens. It fixed that without cutting anyone off.
An analysis of Rippling's approach to measuring AI spend, routing work, and judging cost against output instead of imposing a blanket budget cap.
AI
The price of AI just fell off a cliff. That is the best news your business got this summer.
A founder's take on AI price competition, open models, and choosing the right model for the work instead of paying for unused capability.
AI
Microsoft is spending $2.5 billion because most AI projects fail. That should tell you something.
What Microsoft's embedded-engineering approach says about AI delivery, and why integration and team adoption matter to smaller businesses.
Security
A vendor says 88 percent of companies had an AI agent security incident. The vendor sells the fix. It is still true.
Examine the sources behind AI agent security claims, then focus on access controls, logging, and the risks of untrusted instructions.
Software
Gartner says AI agents put $234 billion in software spending at risk. Read the fine print before you cancel anything.
A closer look at Gartner's SaaS forecast, the limits of agentic AI, and how to decide whether a subscription is worth replacing.
Business
The AI Gold Rush Has a Customer Problem. You Are Not Invited.
A founder's view of enterprise AI investment and what smaller businesses can learn about choosing useful projects and paying for implementation.
General
Welcome to the Mojo AI Services blog.
Notes from founder Joey Epstein on practical AI, software development, and the decisions behind useful business technology.
Tell us what's on your mind.
You don't need a polished brief to reach out. A two-line email about what's bugging you is plenty; we'll tell you straight if we're the right fit, and what we'd tackle first.
We'll scope the work around your workflow, goals, and timeline before quoting anything, so you know what's included before committing.