Loading…
Loading…
547field notes · new posts weekly
Claude Code, agent workflows, Next.js and India finance. Every post is checked against the live tool or the current spec before it ships.
We only write about what we deployed.
We only link to what we bought.
We name names when something sucks.
547 notes
Claude Haiku 5.5 launched on 7 October 2026 with a two-tier price that jumps fivefold when a prompt passes 100K tokens. This post works through a budget for one million agent calls, compares Haiku 4.5 and Gemini 3.5 Flash, and shows how to keep loops in the cheap tier.
Read →Codex has no price per task, only a rolling 5-hour window and a weekly cap inside ChatGPT plans from $0 to $500. This post turns the October message estimates into a planner for a solo developer and a five-person team, and shows when the API-key path or Pro 500 Ultrafast makes sense.
Read →October 2026 packs 13 GST and income tax due dates, and CBDT Circular 07/2026 moved the tax audit report date to 21 October for audit cases. This post lists every date with who it applies to, the late-fee exposure and how to export the dates to a calendar.
Read →A freelancer billing 18 lakh can owe nothing while one billing 48 lakh owes about 3.12 lakh, depending on four decisions. This guide covers 44ADA, books, GST registration, export of services with a letter of undertaking, TDS reconciliation, advance tax and the October 2026 dates, with two worked examples.
Read →Vercel Labs released scriptc on 22 July 2026, an experimental compiler that turns TypeScript into native executables with no JavaScript engine in the default tier. Startup drops to 1.78 ms and memory to a few MiB, while compute-heavy code runs roughly 7.5 times slower than Node 24. This post works out the break-even point and tells you which workloads belong on it.
Read →At DevDay on 29 September 2026 OpenAI introduced Dots, persistent agents powered by GPT-6 Astra that keep working after the chat ends. Chatting with a Dot is free of allowance, but the work it delegates is not. This post covers the approval model, five documented limitations and a budgeting method that keeps a Dot from draining your Codex quota.
Read →GPT-6.1 Sol launched with one-fifth the token prices of GPT-6 Astra: $2.00 and $10.00 against $10 and $50 per million tokens. This post prices three realistic agent runs at 1,000 repetitions, adds the cached-input and Ultrafast tiers, and shows the one condition under which paying five times more for Astra is still the cheaper choice.
Read →Polars 2.0 changes the default engine for LazyFrame.collect() from in-memory to streaming, and streaming does not promise row order for group_by, joins or unpivot. This guide walks through eight breaking changes that can alter pipeline output without raising an error, each with before and after code taken from the official upgrade guide.
Read →Mistral Large 4 entered public preview on 6 October 2026 as a roughly 1 trillion parameter mixture-of-experts model with open weights promised by the end of the month. Two different prices circulate, and vendor and independent benchmarks tell different stories. This post separates the claims, prices a standard agent run against Haiku 5.5, GPT-6.1 Sol and Gemini 3.5 Flash, and says who should wait for the weights.
Read →Spotify did not ban AI music. It banished synthetic identities. From mid-September 2026, profiles that look like AI-generated artists get an AI Persona badge and leave editorial and algorithmic recommendations unless a listener already follows them. The badge is about the face on the profile, not the tools behind the track. This is the identity, disclosure and catalogue playbook for AI-assisted artists who want to stay in the pipes.
Read →A July 2026 Nature Machine Intelligence study held prompts, tools and compute constant and varied only coordination structure. Multi-agent architectures burned 58% to 515% more reasoning turns than a single agent and, at equal token budgets, did not reason better. For a solo builder that is a tax, not a feature. This is the one-agent default, the three tests that justify a second agent, and the lock-and-gate pattern that replaces crew roleplay.
Read →Character drift is the most common failed promise in generative video, and more reference images do not fix it. What fixes it is a production system: a hero sheet that separates immutable from mutable traits, a two-axis minimum lock, a continuity card per shot, and a drift score that tells you when to stop regenerating and fix the document. Model-agnostic, and it holds a face across a 6 to 8 shot sequence.
Read →Monday memo · free
1 field report, 3 links we stole ideas from, and 1 product we actually bought. No spam, unsubscribe in one click.
Beyond the writing
4,300+ developer tools, templates and starter kits — plus every free browser tool, no signup.
Full archive