[ Blog_Posts ]

Pi Ships Zero Lines of MCP. Codex Ships 53,713.

One harness treats MCP as core infrastructure and spends tens of thousands of lines on it. The other refuses it on principle and tells the agent to write its own tools instead. Both positions are coherent, and the disagreement is not about capability. It is about who owns the schema, what a stable boundary is worth, and what you pay for it in tokens and attack surface on every single request.

2026-09-18 Harrison Guo 8 min read mcp model-context-protocol openai-codex pi-agent tool-schema agent-architecture extensibility agent-security

Your Agent Session Is a Tree. Your Code Thinks It Is a List.

Codex made Item a wire type and thread/fork a protocol call. Pi gave every session entry a parentId. Claude Code compresses the transcript in five stages. Three harnesses, three answers to what the durable unit of agent state actually is, and the answer decides whether running out of context costs you a summary or costs you the work.

2026-09-15 Harrison Guo 8 min read agent-state session-management context-compaction openai-codex pi-agent claude-code agent-architecture cache-coherence

550 Tokens of Prompt, 588 Tokens of Tools

Pi is famous for a system prompt under a thousand tokens. I measured it: 550. Then I measured the four tool schemas that ship beside it in every single request: 588. The celebrated number is the smaller half. Meanwhile Codex ships a different prompt length for every model, from 1,436 to 5,070 tokens, and the reason why is the most useful thing in either codebase.

2026-09-11 Harrison Guo 8 min read system-prompt token-budget pi-agent openai-codex prompt-engineering cost-curve agent-architecture ai-cost

Same Model, 13.3% to 38.3%

OpenAI changed two API settings on ARC-AGI-3 and GPT-5.6 Sol went from 13.3% to 38.3% while spending six times fewer output tokens. Same model, same weights, same benchmark. The official harness had been throwing away the model's reasoning after every move. Which means the score was never a measurement of the model, and neither is most of what we compare.

2026-09-08 Harrison Guo 9 min read arc-agi-3 benchmarks agent-harness evaluation gpt-5 context-engineering ai-engineering measurement

Building a Bootloader from Scratch — Just Assembly, No OS

512 bytes, no operating system, no standard library, no memory management — BIOS drops you at 0x7c00 in 16-bit real mode and everything after that is yours to build. A hand-written Stage-1 MBR, line by line, and why the exercise surfaces every abstraction you normally stand on.

2026-09-07 Harrison Guo 5 min read bootloader assembly x86 bios mbr low-level os-development

Debugging a Real-Mode Bootloader in GDB — When CS Changes, Symbols Break

Step through a hand-written x86 bootloader in GDB, hit a far jump, and watch your function names turn to garbage. Nothing crashed — CS changed, and in real mode that moves every address out from under GDB's symbol table. Why it happens and how to keep symbols following you.

2026-09-07 Harrison Guo 5 min read bootloader x86 assembly gdb real-mode debugging low-level

Debugging the Ubuntu 6.8 Kernel in GDB & QEMU — Without Rebuilding It

Attach GDB to a stock Ubuntu 6.8 x86-64 kernel in QEMU, set a breakpoint, and watch nothing happen. The kernel is running fine. The bug isn't in the code — it's that KASLR moved the kernel and your symbols didn't. How to fix it without a rebuild.

2026-09-07 Harrison Guo 4 min read kernel x86_64 ubuntu qemu gdb kaslr linux debugging

Building & Debugging a Custom ARM64 Linux Kernel — Yocto, QEMU, GDB

Rebuild an ARM64 Linux kernel with Yocto, boot it in QEMU, and step through it in GDB — on a laptop, no hardware. The workflow is the easy part; the part nobody tells you is why GDB can't find your source and how to fix it.

2026-09-06 Harrison Guo 5 min read kernel arm64 yocto qemu gdb linux debugging embedded

Cache Miss, TLB Miss, False Sharing — Three Killers Your Profiler Won't Name

top shows 100% CPU. perf shows the truth. Two threads doing identical work run in 841ms or 8118ms depending on one thing your profile never mentions: where their data sits in memory. Cache misses, TLB misses, and false sharing, with real code and real numbers.

2026-09-06 Harrison Guo 6 min read cpu cache tlb false-sharing perf performance memory-model coretracer

eBPF on Linux — kprobe vs fentry: Hooking Internals & What Production Observability Misses

kprobe and fentry both hook kernel functions from eBPF, but they install differently, cost differently, and hand you arguments differently. A source-level comparison, grounded in a real eBPF project, plus why the choice is a production cost — not trivia.

2026-09-06 Harrison Guo 9 min read ebpf kernel linux observability kprobe fentry tracing bpf

[ Blog_Posts ]

[ Connect_With_Me ]