Achint Mehta is a senior software development manager and AI security researcher — writing here about AI agents, low-level systems, and computer science. More about me
The Latest
Using a spec as a benchmark: What 90 model runs taught me about agents, prompts, and shortcuts
I ran the same OpenSpec through 90 runs across models, agents, and configurations. Some results were expected. Others were not.
AI and the changing structure of work
A conversation with Deepak Tripathi, who leads a Global Capability Center, challenged several of my assumptions about how organizations develop capability. Here is what the research says about where AI is actually putting the pressure.
If AI is replacing engineers, why are companies still hiring them?
The ATM was supposed to kill bank tellers. It didn't. Understanding why tells you more about AI and software engineering than most takes you'll read on LinkedIn.
Introduction to AI Agents
What AI agents are, why they exist, and how to set up a local environment to start building them with LangChain and LangGraph.
02LLM Clients and Chains
How to talk to LLMs from code, ChatOpenAI vs OpenAI, composing chains with LCEL, and the invoke, batch, and stream invocation patterns.
03Chunking Strategies
How to break documents into retrieval-friendly chunks, fixed-size, structure-aware, language-aware splitting, and choosing the right chunk size.
One Number Is Not a Finding
Why "the model scored 74" tells you almost nothing on its own, what runs, conditions and cells are, how much the same experiment moves when you simply run it again, and the three questions every result has to answer before it counts as a finding. With the twenty-four runs that the rest of the series will keep coming back to.
02Middles, Spreads and Rates
The mean and the median, and the runaway token cost that pulls them apart. Trimmed, weighted and geometric means and when each one is honest. Standard deviation, the interquartile range and how to read a box plot. Then rates, which are averages in disguise, and the four ways of putting an interval on "16 out of 18", one of which claims a pass rate of 103 per cent.
03The Claim and Its Shadow
What a hypothesis is and the four things it has to contain, why every hypothesis comes with a null hypothesis attached and why it is the null that gets tested, one-sided and two-sided claims, primary and secondary contrasts, and the six hypotheses from the verification study written out as bets, three of which lost.
-
Type Qualifiers
The three C type qualifiers, const, volatile, and restrict, what they mean, and how to apply them to variables, pointers, and objects.
-
RFC
What an RFC is, how the RFC Editor works, the types of RFCs, the Internet standards track, and the tradition of joke RFCs.
-
Processor Modes
Why grouping similar CPU instructions together improves performance by reducing mode-switching overhead, illustrated with the Pentium MMX example.
-
Null Pointer
Why NULL is reserved for pointer operations in C, how it is defined, and why mixing it with integer zero causes problems.
-
Magic Numbers
Why hardcoded numeric literals in code are harmful, and how to replace them with named constants, macros, enumerations, or typed constants.
-
GDB step-by-step debugging
A practical walkthrough of debugging C programs with GDB, breakpoints, watchpoints, and reading stack frames.
-
What is GCC
The history of GCC, from Richard Stallman's original release to the EGCS merger, GPL licensing, and Windows support.
-
Using GCC
How to compile single and multi-source C programs with GCC, object files, the -g flag, -c flag, -I flag, and PATH setup.
-
Static Library
How to create and use static libraries in C, libraries vs object files, ar, ranlib, naming conventions, and link order.
IBM 701, The Speed Demon
IBM's first commercially available scientific computer, introduced in 1953 and born from the demands of the Korean War.
02Fortran (1957)
The world's first high-level programming language, created by John Backus at IBM to make programming accessible beyond assembly.
03COBOL (1959)
The business programming language born from a Pentagon-led collective effort, designed with "maximum use of Simple English" in mind.