suhailmohebi
Field notes
An open ivory folio with a crystal sphere and glass prism resting on its pages, on a green marble desk
Conceptual illustration

// 48 DEEP DIVES · FOUR READING LANES

Find the method for the problem in front of you.

Start with the job: operate, verify, ship, or explore. Every path returns to evidence and an explicit next move.

01 / OPERATE

Direct the work

Tools, settings, voice, control surfaces, and the contracts that keep agent work bounded.

An orderly wooden desk with open paper manuals, filed binders and a brass lamp beside a window43 · OPERATIONS Running AI Agents in Production: An Operations ManualScoping, the authority line, evidence ladders, honest monitors, durable memory, and fleet coordination — one connected manual. A compass, a magnifying glass, a leather notebook and a brass ruler laid out on a green marble bench01 · TOOLS The four-agent toolbenchClaude, Codex, Droid and Cursor — what each is best at, how to route work between them, and the one mirrored context file that keeps them in sync. Close-up of a knurled copper rotary dial with a fine engraved scale02 · SETTINGS Tuning the knobs: models & settingsReasoning effort, model tier, temperature, tool-use — and separate saved profiles for coding speed versus agentic work, with copy-usable configs. A vintage-style studio microphone beside a blank sheet of paper on a marble desk03 · WORKFLOW Voice-driven developmentDriving a fleet of agents by talking — handling transcription artifacts, and the shorthand that expands one spoken phrase into a whole workflow. A smartphone resting on an ivory writing pad beside a copper pen17 · OPERATIONS Hermes Agent: Telegram & OpenRouter setup guideWhat Hermes is, how to install it, connect a private Telegram bot, switch models, add skills and scheduled work, and keep the assistant safe. A closed copper ring with a small glass inspection window, standing on a marble desk18 · SYSTEMS Designing skills and loops for certain and uncertain outcomesHow to package agent skills as explicit contracts, use deterministic loops for exact predicates, and bound non-deterministic work with rubrics, budgets, reviewers, and human gates. A dark round instrument face with a single small amber light, on a brass stand20 · OPERATIONS The quiet monitorA monitoring architecture that preserves raw observations, distinguishes unknown from unchanged, and interrupts only when a decision threshold is crossed. A smartphone with a blank screen beside an open notebook and pen on a marble desk22 · OPERATIONS Codex in my pocketHow I start and steer Codex work from my phone, keep a Mac available, and return to verified results. A small vented server case standing on a green marble plinth in a study23 · INFRASTRUCTURE How I run AI agents on a Hetzner VPSCreate and secure the server, install the runtime, connect Telegram, isolate multiple agents, and operate the system with evidence. A copper drafting frame holding a single clean sheet of ivory paper24 · SCOPING How to scope an AI agent taskA five-part method for turning an ambiguous request into a bounded task with a clear outcome, authority, evidence, and stop conditions. An overstuffed stack of loose papers beside one thin, neatly squared brief on a marble desk25 · PERFORMANCE Why my new AI agent was slow before it answeredA measured AI-agent prompt-bloat teardown of fixed context, skill indexes, tool schemas, and the profile rules that separate prompt weight from end-to-end latency. A copper button under a clear protective cover on a marble base32 · AUTONOMY The agent never touches the buttonAn automated research fleet that proposes, tests, and verifies everything — governed by one file and adversarial review — while a human carries every irreversible action the last step.

02 / VERIFY

Protect the truth

Source-of-truth checks, evidence loops, privacy boundaries, and failures that looked successful.

A brass gauge laid across the edge of a thick stack of paper30 · VERIFICATION Policy as a testEncode a direction-level decision as an automated check so later sessions cannot quietly undo it—and name where the technique stops working. A microscope aimed at an empty specimen mount on a marble desk28 · FAILURE When the monitor’s subject diesAssert that the subject still exists before a vacuous pass can be mistaken for a healthy system. A loose sheet set aside next to an open book whose page still shows the faint impression of it27 · FAILURE The page you deleted is still being servedClose the origin, feed, index, and external-verification loop after retiring a public page. A brass and glass inspection bar with five round lenses, four lit and one dark26 · SECURITY How I built a fail-closed vulnerability watcherFailed feeds, uncertain product matches, and unknown versions remain evidence gaps—not reassuring conclusions. A brass lamp shines a beam through three clear glass prisms onto an ivory card on a marble desk06 · METHOD The verification loopA practical evidence ladder for turning ‘looks right’ into a result another person or agent can trust. Layered ivory paper screens with a single square opening revealing the view beyond07 · CONTEXT Context without oversharingHow to give agents enough context to work well while keeping private systems, personal details, credentials, and non-public facts out of public artifacts. A pressed leaf under a glass plate beside two paper prints of the same leaf on a green marble desk09 · ORIENTATION Before the AI edits: find the live truthA source-of-truth protocol for checking production, deploy state, repositories and generated artifacts before an agent changes the wrong version. A glowing bulb on a brass mount beside a mechanical linkage that is visibly disconnected13 · FAILURE When ‘success’ is still brokenWhy a 200 response, green check, or finished-looking agent answer can still hide a failed user outcome—and what evidence closes the gap. Two nearly identical prints of a tree side by side, one slightly out of register14 · FAILURE The stale worktree that passed every checkA postmortem pattern for the dangerous case where tests are green because they ran against the wrong source of truth. A brass magnifying glass resting on layered archival papers and photographs19 · EVIDENCE Provenance before conclusionsHow to turn scattered public data into a reproducible evidence archive where every file, limitation, and claim can be traced. A camera lens lying on its side facing a small round mirror on a stand21 · DIAGNOSIS Diagnose the observer firstA layered incident method for separating device, network, routing, remote-system, and user-outcome failures before changing anything. A narrow band of daylight falling through the gap between two dark marble wallsFIELD NOTE · SECURITY AI is collapsing the CVE response windowWhy vulnerability discovery, disclosure, and weaponisation now demand a faster intelligence loop with human-owned decisions.

03 / SHIP

Make the work compound

Recipes, publishing, handoffs, coordination, and memory that improve the next run.

A strip of film unwinding from a brass reel across a marble desk, with a brass loupe over one frame45 · VIDEODirecting AI video with HiggsfieldFrom prompt-only concepts to reference-driven sequences: shot timing, voice continuity, motion constraints, and a 15-job Seedance 2.5 experiment. Two stacks of archive papers joined by a long ribbon of paper on a marble desk29 · RECIPES The branch that predates the decisionSweep a whole long-lived branch for retired assumptions before a tidy merge silently restores them. An open folio with blank pages, copper clips and index tabs on a marble kitchen counter04 · RECIPES The cookbookSeven copy-usable recipes — cache-bust blast radius, CSP hash recompute, adversarial verify, config recovery from live headers — each with the reasoning. Loose blank sheets on the left giving way to a neat stack and a bound green book on the right08 · WORKFLOW From rough brief to published pageHow a loose idea becomes a structured, visual, connected page—and leaves the seed for what to publish next. An open green envelope holding neatly arranged blank ivory sheets10 · HANDOFF Handoff packets that do not lose the plotA copy-usable transfer format for current state, decisions, evidence, stop conditions, and the smallest safe continuation. Several copper tracks converging into one at a single junction on a marble desk11 · COORDINATION Parallel agents without branch collisionsA practical ownership and integration method for running agents in parallel without colliding on shared files, generated artifacts, cache versions, or release state. A thick green leather-bound ledger with blank index tabs on a marble desk12 · MEMORY Decision ledgers for long-running agent workA practical way to preserve why an agent rule exists, append evidence that confirms or changes it, and name the trigger for its next review.

04 / EXPLORE

Local AI: models, memory and performance

Choose a Mac model, compare inference hardware, or plan an RTX 5090 serving setup. Start with the Qwen3.8 Mac memory guide and the SGLang vs vLLM profile comparison.

Brass blocks of different sizes packed into an open travertine stone vessel on a green marble desk, filling it exactly to an engraved line inside the rim, with one small brass block left over beside it48 · LOCAL AIWill this model fit? A local LLM memory calculatorPick a GPU or Mac, a model, a quantization and a context: an estimate of weights, KV cache and overhead, and whether it fits.Two small brass engines on wooden plinths on a green marble desk, the right one driving a second, belt-linked wheel that spins beside its flywheel47 · LOCAL AISGLang vs vLLM, measured on one RTX 5090SGLang with a drafter against vLLM with MTP or none, on one 32 GB card: generation, prompt speed, memory and stability, measured.Two identical brass oil lanterns burning with the same flame on a green marble desk, the right one with its glass chimney lifted off and lying beside it46 · LOCAL AIPopular local models vs abliterated versionsThe most downloaded Hugging Face models for an M4 Max, M5 Max or DGX Spark, what fits in 128 GB, and how abliterated versions really differ.A small glowing core at the centre of a large nested marble and stone cube44 · HARDWAREQwen3.8-Flash-Next system requirementsCan Qwen3.8-Flash-Next run on an RTX 5090 or a Mac? A source-checked guide to NVFP4 memory, CPU offload, runtime support, and the 6B-active trap. Close-up of a graphics card's metal heat-sink fins and copper heat pipes in raking light42 · LOCAL AI Serving LLMs on an RTX 5090: The Working RecipesFour validated profiles from 84.5 to 139.7 tok/s, the three context numbers that matter, and the traps that cost decode. A compact unbranded aluminium desktop computer beside a tall stack of translucent glass panes41 · LOCAL AI Local LLMs on Apple Silicon: The Field GuideWhich Mac runs what, 4-bit or 8-bit, the measured speculation gains, KV budgets, and when the GPU box wins. Two stacks of clear glass plates, one short and one tall, side by side on a marble desk40 · LOCAL The other 4-bit decision4-bit weights were nearly free; a 4-bit KV cache halves reasoning scores. The per-token math, the mechanism, and the budget a 32 GB card forces. A brushed aluminium block beside a black finned heat sink on a marble slab39 · LOCAL Same model, two machinesThe 27B on both a 128 GB M4 Max and a 32 GB RTX 5090: 46× apart in prefill, 5× in decode, 3× with speculation — and what is genuinely not comparable. A long zigzag-folded paper ribbon stretching from a copper holder across a desk38 · LOCAL The rule was 250,000 tokens agoA 40-check graded suite at four depths of a 262k window — tool calls clean everywhere, one trailing period wasn’t, and the grader failed before the model did. Two parallel copper tracks curving across a marble desk, a finer guide rail running ahead of the main track37 · LOCAL Speculative decoding, measured on my M4 MaxNative MTP +24%, a DFlash 2 drafter up to +183%, the acceptance rates behind them, and the throughput metric that nearly reverted a working config. Three carved stone owls of the same shape, each with a different level of detail, from blocky to finely carved36 · LOCAL AI How I Choose a Quantization for Local InferenceThe formats that matter, the 32 GB memory math, where quality actually drops, and the nine-point test that decides Q4, Q8 or NVFP4. Two lamp-lit worktables in separate rooms, joined by an arched marble passage35 · LOCAL AI I Set Up My RTX 5090 AI Workstation Without Touching ItQwen 3.8 27B across vLLM, SGLang, context and KV-cache tests—then a private off-site inference proof through Tailscale. A triple-fan graphics card on a green marble desk in a home study overlooking hills and a lake34 · ARTIFICIAL INTELLIGENCE Building My Personal AI Lab: RTX 5090, Qwen3.6 and Local InferenceA dedicated local-compute plan for MTP inference, agentic workflows and reproducible open-weight model benchmarks. An unbranded compute board inside a clear acrylic case on a marble desk05 · LOCAL Local & uncensored modelsRunning open-weight models on your own machine — the stack, why privacy and cost make it worth it, and a plain, responsible explanation of what abliteration is. A closed green field notebook beside a small rough mineral specimen15 · LOCAL Local model field notesA living test contract for local models — runner, quantization, configured context, latency, failure modes, and honest cloud comparisons. A brass balance scale with graduated brass weights on a marble base16 · RISK Position sizing for agent autonomyWhat derivatives, portfolio construction, and surviving volatility teach about permissions, stop conditions, and bounded agent action. An ivory bound ledger beside a set of brass weights, with a city skyline beyond the window31 · MARKETS Building a paper-first DFM trading systemHow an automated DFM and ADX research system handles liquidity, two-tier allocation, cash-aware sizing, manual execution, and backtests that are allowed to fail. A neat stack of clear glass plates on a broad brushed-aluminium surface33 · LOCAL Qwen 3.8 moved the frontier onto my MacThe ELI5 Opus 4.6 comparison, useful local jobs, and a practical guide to choosing 4-bit or 8-bit for your Mac's RAM.

// CUSTOM SKILLS

Turn a solved problem into callable work.

Custom skills package a method as either a repeatable function or a bounded loop—with named inputs, evidence, limits, output, and an exit condition.

/seo-cluster-audit/release-verification/handoff-packet
Explore the custom skills

// READ → OPERATE

Turn a useful idea into a repeatable move.

The templates carry the evidence requirements into your own context; the Now page shows what is currently under test.

Use the templates See what is running