AI Productivity

Using AI coding assistants well: prompting, review habits, and where they genuinely help versus where they don't.

All posts

A dark comparison-matrix style cover reading Copilot's Annual Multiplier Spike, badged legacy multiplier table revised June 1 2026, listing five row summaries: GPT-5.5 sits at 57x the table's real ceiling, Claude Opus 4.6 through 4.8 all carry a 27x multiplier, Pro+ annual gets 1,500 premium requests and no more, legacy annual plans get no new models or features, and subscribers can stay cancel for refund or upgrade to Credits

Annual Copilot Plans: Model Multipliers Just Spiked

GitHub revised Copilot's legacy model-multiplier table for annual Pro/Pro+ plans on June 1, 2026, some models jumped as high as 57x.

Aug 19, 2026Read more →

AI Productivity

Check Your Copilot AI Credits Usage Before Sept 1

See GitHub Copilot AI Credits usage per cycle, per user via API, and org-wide before the Sept 1, 2026 allowance cut hits your bill.

Aug 19, 2026Read more →

AI Productivity

Why Copilot Agent Mode Burns Through Credits So Fast

GitHub Copilot's Agent Mode spends AI credits per tool-calling step, not per message. Here's the actual mechanism, with real token math and a sourced case.

Aug 19, 2026Read more →

A dark comparison-matrix style cover reading GitHub Copilot's AI Credits Cliff, badged Sept 1 2026 promo credits revert, listing five row summaries: Business credits dropping 3,000 to 1,900 per user per month, Enterprise dropping 7,000 to 3,900, 1 credit equals $0.01 priced per token per model, annual plans on a separate legacy multiplier system, and AI credits paid usage enabled by default so metered billing continues uncapped unless an admin disables it

AI Productivity

GitHub Copilot's AI Credits Cliff: Full Guide

GitHub Copilot's promotional AI Credits allowances revert to standard levels on September 1, 2026. What changed June 1, what reverts, and how to prepare.

Aug 19, 2026Read more →

AI Productivity

How GitHub Copilot AI Credits Are Actually Priced

GitHub Copilot's AI Credits convert real per-token, per-model pricing into dollars at a fixed 1 credit = $0.01 rate. Here's exactly how the math works.

Aug 19, 2026Read more →

AI Productivity

Setting Copilot Budget Controls Before the Sept Cliff

Configure GitHub Copilot AI credits budgets before Sept 1, 2026: set the org/cost-center stop-usage toggle and per-user hard stop limits.

Aug 19, 2026Read more →

A dark comparison-matrix style cover reading GitHub Copilot Plans Compared by AI Credits, badged credits-per-dollar math verified 2026-08-19, listing six row summaries: Pro at $10 per month for 1,500 credits or 150 per dollar, Pro+ at $39 for 7,000 credits or about 180 per dollar, Max at $100 for 20,000 credits or 200 per dollar, Business at $19 per seat for 1,900 credits or 100 per dollar, Enterprise at $39 per seat for 3,900 credits or 100 per dollar, and Free with no published credit figure capped at 2,000 completions per month

AI Productivity

GitHub Copilot Plans Compared by AI Credits, Not Price

GitHub Copilot plans don't scale AI credits with price. Real credits-per-dollar math across Free, Pro, Pro+, Max, Business, and Enterprise.

Aug 19, 2026Read more →

A dark comparison-matrix style cover reading GitHub Copilot's PRU to AI Credits Migration, badged June 1 2026 billing cutover, listing five row summaries: PRU's flat per-request multiplier removed for monthly plans, credits price real input output and cached tokens per model, premium requests first enforced June 18 2025, annual Pro and Pro+ subscribers stayed on PRU not migrated, and a lapsed annual plan downgrades to Free not to Credits

AI Productivity

GitHub Copilot's PRU to AI Credits Migration, Explained

GitHub retired Premium Request Units for AI Credits on June 1, 2026, for monthly Copilot plans. Here's what changed under the hood, and what didn't.

Aug 19, 2026Read more →

A dark comparison-matrix style cover reading Copilot's Sept 1 Credit Cliff Checklist, badged 4-step checklist for the Sept 1 2026 deadline, listing four row summaries: check usage across settings, API, and dashboard; decide the AI credits paid usage policy on purpose; set the universal user-level budget first; and re-size every budget to the 1,900/3,900 standard pool

AI Productivity

What to Do Before GitHub Copilot's Sept 1 Credit Cliff

A step-by-step checklist to check Copilot AI credit usage, set budget controls, and size your budget before GitHub's Sept 1, 2026 credit cliff hits.

Aug 19, 2026Read more →

AI Productivity

Pick Cheaper Copilot Models to Stretch Your Credits

GitHub Copilot prices every model differently. Here's when a cheaper model is genuinely enough, and when frontier capability is worth the extra credits.

Aug 19, 2026Read more →

A dark comparison-matrix style cover reading When Copilot Credits Run Out, badged Default behavior verified Aug 2026, listing six row summaries: no budget bills overage automatically, disabling paid usage blocks at zero, a budget with Stop usage off keeps accruing charges, a budget with Stop usage on blocks at the limit, user-level budgets always hard-stop with no toggle, and individual Pro Pro+ Max plans never bill silently

AI Productivity

What Happens When GitHub Copilot Credits Run Out

GitHub Copilot's paid overage is on by default for orgs once AI credits run out - here is exactly what happens with and without a budget.

Aug 19, 2026Read more →

A dark comparison-matrix style cover reading The 2026 LLM Token and Pricing Reset, a 10-part series on GPT-5.6, Claude, and Gemini, listing four row summaries: GPT-5.6's bigger window and pricier caching, Claude Opus 4.8's unchanged price but heavier tokenizer, Gemini 3.6 Flash's fewer tokens and new price, and legacy GPT-4/o1/o4-mini shutting down October 23 2026

AI Productivity

The 2026 LLM Token & Pricing Reset: Full Guide

GPT-5.6, Claude Opus 4.8, and Gemini 3.6 all changed token limits and pricing in 2026. Full breakdown of what changed, and how to check your own numbers.

Aug 12, 2026Read more →

AI Productivity

Claude Opus 4.7 Breaks temperature and top_p Params

Claude Opus 4.7 returns HTTP 400 for temperature, top_p, top_k, and budget_tokens in some configurations. What changed and how to fix it.

Aug 12, 2026Read more →

AI Productivity

Claude Opus 4.8: Same Price, Cheaper Fast Mode

Claude Opus 4.8 kept base pricing at $5/$25 per MTok but cut Fast Mode from $30/$150 to $10/$50. What that means for latency-sensitive calls.

Aug 12, 2026Read more →

AI Productivity

Claude's New Tokenizer Counts Up to 35% More

Claude Opus 4.7's tokenizer reportedly counts up to 35% more tokens for the same text. Anthropic's own GA changelog doesn't confirm it.

Aug 12, 2026Read more →

AI Productivity

Gemini 3.6 Flash Cuts Output Tokens 17%, Price Too

Gemini 3.6 Flash uses 17% fewer output tokens than 3.5 Flash and costs $0.75/$3.75 per million tokens through 2026, rising to $1.50/$7.50 in 2027.

Aug 12, 2026Read more →

A terminal window showing OpenAI's own deprecations page: generally available models get at least 6 months notice before retirement, and gpt-4, gpt-4-0613, o1, o1-pro, and o4-mini all shut down 2026-10-23, replaced by gpt-5.6-sol and gpt-5.6-terra

AI Productivity

gpt-4, o1, and o4-mini Shut Down October 23, 2026

OpenAI's gpt-4, o1, and o4-mini model IDs stop responding on October 23, 2026. The exact deprecation notice, and how to migrate to GPT-5.6.

Aug 12, 2026Read more →

A dark comparison-matrix style cover reading GPT-5.6's Context Window, 922K input and 128K output tokens on o200k_base, listing three row summaries: old context window 128K to 200K tokens, new context window about 1.05M tokens, and this site's token counter not yet supporting o200k_base

AI Productivity

GPT-5.6's Context Window: 922K In, 128K Out Tokens

GPT-5.6 shipped a ~1.05M-token context window (922K input, 128K output) on the o200k_base encoding. What changes for existing integrations.

Aug 12, 2026Read more →

A dark comparison-matrix style cover reading GPT-5.6 Ends Free Prompt-Cache Writes, a 1.25x premium with a 30-minute minimum TTL, listing three row summaries: before automatic free caching, after explicit breakpoints with a 1.25x write cost, and cache reads staying at roughly 90% off unchanged

AI Productivity

GPT-5.6 Ends Free Prompt-Cache Writes (1.25x Premium)

GPT-5.6 replaced OpenAI's free implicit prompt caching with explicit breakpoints and a 1.25x write premium. What changes in your integration.

Aug 12, 2026Read more →

AI Productivity

OpenAI Extends Prompt Cache Retention to 24 Hours

OpenAI extended prompt cache retention from a few minutes to up to 24 hours. What that changes for agents and pipelines with repeated calls.

Aug 12, 2026Read more →

AI Productivity

OpenAI vs Claude: Prompt Caching Cost Math in 2026

OpenAI and Anthropic both charge a 1.25x premium on cache writes; Anthropic's rises to 2x for its 1-hour tier. How the real costs compare.

Aug 12, 2026Read more →

A dark comparison-matrix style cover reading Same Prompt, Different Bill, GPT-5.6 vs Claude Opus 4.8 vs Gemini 3.6, listing three row summaries: GPT-5.6's 922K-token window and pricier caching, Claude Opus 4.8's unchanged price but heavier tokenizer, and Gemini 3.6 Flash's 17% output-token cut and new price

AI Productivity

Same Prompt, Different Bill: GPT-5.6 vs Claude vs Gemini

GPT-5.6, Claude Opus 4.8, and Gemini 3.6 all changed token counts and per-token pricing in 2026. See what shifted, and check your own prompt free.

Aug 11, 2026Read more →

A dark comparison-matrix style cover listing the MCP 2026-07-28 spec rewrite's changes: Mcp-Session-Id removed, initialize handshake replaced by _meta fields, server/discover now required, and SSE subscriptions replaced

AI Productivity

The MCP 2026-07-28 Spec Rewrite: Full Guide

The MCP 2026-07-28 spec rewrites the protocol core: sessions gone, a new required RPC, subscriptions replaced. Full rundown of every breaking change.

Aug 6, 2026Read more →

A terminal window showing the verbatim MCP 2026-07-28 changelog entry: OAuth 2.0 Dynamic Client Registration Protocol deprecated in favor of Client ID Metadata Documents

AI Productivity

MCP Deprecates OAuth Dynamic Client Registration

The 2026-07-28 MCP spec deprecates OAuth DCR for Client ID Metadata Documents. A stable HTTPS URL becomes the client_id instead of a minted registration.

Aug 6, 2026Read more →

A terminal window showing the verbatim MCP 2026-07-28 changelog entry: Roots, Sampling, and Logging features deprecated, suggested migrations provided

AI Productivity

MCP Deprecates Roots, Sampling, and Logging

The 2026-07-28 MCP spec deprecates Roots, Sampling, and Logging with a 12-month runway. Each has a real migration path: config, MRTR, and stderr/OTel.

Aug 6, 2026Read more →

A terminal window showing the verbatim MCP 2026-07-28 changelog entry: moved experimental tasks to official extension io.modelcontextprotocol/tasks, replaced blocking tasks/result with polling

AI Productivity

MCP Moves Tasks Out of Core Into an Extension

The 2026-07-28 MCP spec moves Tasks into io.modelcontextprotocol/tasks and replaces blocking tasks/result with polling via tasks/get and tasks/update.

Aug 6, 2026Read more →

A terminal window showing the verbatim MCP 2026-07-28 changelog entry: removed protocols ping, logging/setLevel, notifications/roots/list_changed

AI Productivity

MCP Removes the ping and logging/setLevel Methods

The 2026-07-28 MCP spec removes ping and logging/setLevel. Log level now sets per-request via _meta, and there is no stated ping replacement.

Aug 6, 2026Read more →

A terminal window showing the verbatim MCP 2026-07-28 changelog entry: eliminated Mcp-Session-Id header from Streamable HTTP transport, servers use explicit handles for cross-call state

AI Productivity

MCP Removes Protocol-Level Sessions (Mcp-Session-Id)

The 2026-07-28 MCP spec eliminates the Mcp-Session-Id header. Cross-call state now needs explicit, server-minted handles passed as tool arguments.

Aug 6, 2026Read more →

A terminal window showing the verbatim MCP 2026-07-28 changelog entry: eliminated Last-Event-ID header and SSE event IDs, broken streams require re-issuing as new requests

AI Productivity

MCP Removes SSE Resumability (Last-Event-ID Gone)

The 2026-07-28 MCP spec eliminates Last-Event-ID and SSE event IDs. A broken stream now requires re-issuing the whole request, not resuming in place.

Aug 6, 2026Read more →

A terminal window showing a real MCP 2026-07-28 request with protocol version and client capabilities carried in the _meta field instead of an initialize handshake

AI Productivity

MCP Replaces initialize Handshake With _meta Fields

The 2026-07-28 MCP spec removes the initialize handshake. Every request now carries protocol version and capabilities in _meta fields instead.

Aug 6, 2026Read more →

A terminal window showing a real MCP notification sent over a subscriptions/listen stream: notifications/tools/list_changed

AI Productivity

MCP Replaces SSE Subscriptions With subscriptions/listen

The 2026-07-28 MCP spec replaces the SSE GET endpoint with an opt-in subscriptions/listen stream. Clients now request each notification type explicitly.

Aug 6, 2026Read more →

A real MCP InputRequiredResult example showing resultType input_required, an inputRequests confirmation elicitation, and an opaque requestState value

AI Productivity

MCP Requires resultType on Every Returned Result

The 2026-07-28 MCP spec makes resultType mandatory on every result. The old server-initiated request pattern is replaced by an input_required result.

Aug 6, 2026Read more →

A terminal window showing the verbatim MCP 2026-07-28 changelog entry: servers must implement server/discover to advertise supported protocol versions, capabilities, and identity

AI Productivity

MCP Servers Must Now Implement server/discover

The 2026-07-28 MCP spec adds a required server/discover RPC. Existing servers need a handler for it before a compliant client can negotiate at all.

Aug 6, 2026Read more →

A person typing on a laptop showing a glowing data readout, in a moody blue-lit setting

AI Productivity

A Practical Prompting Guide for AI Coding Assistants

Concrete techniques for prompting AI coding assistants: specificity, constraints, checkpoints, and reviewing diffs, for working code on the first pass.

Jul 27, 2026Read more →