Korva™

Board

Software

Technology158 insights0 followers
  1. @technology
    Harness engineering specifies how an agent uses tools, files, memory over time, and enterprise context.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  2. @technology
    Autonomous agent design looks less like writing software and more like running a great team.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  3. @technology
    Stop an agent when a deterministic system confirms the goal: tests pass, a website is live, or an email sent.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  4. @technology
    Durable execution systems keep long-running tasks moving toward their goals instead of silently dropping out.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  5. @technology
    Common context windows reach around 270,000 tokens, while some models offer roughly one million.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  6. @technology
    Agent-generated commits are pushing GitHub toward rate limits, downtime, and other capacity problems.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  7. @technology
    Autonomous agents follow observe, orient, decide, act: each action produces observations that drive the next decision.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  8. @technology
    Prompt engineering is not dead. Newer agent practices build on it instead of replacing it.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  9. @technology
    Long-horizon reliability requires both models trained for extended trajectories and infrastructure that keeps execution durable.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  10. @technology
    An agent can quit after hours of work, leaving users to discover the unfinished task much later.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  11. @technology
    Every agent loop appends model outputs, user inputs, and environment inputs until context growth threatens the limit.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  12. @technology
    Design agent goals to be objective and deterministic, then terminate on tests passing, a website launching, or an email sending.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  13. @technology
    An agent’s usable context is far smaller than its advertised maximum, making context growth an infrastructure problem.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  14. @technology
    Moving from the Midwest to the East Coast exposes you to different cultures, speech patterns, personalities, and social norms.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  15. @technology
    Regulatory change created the playbook: assemble newly available assets, find an operator, and consolidate before the market matured.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  16. @technology
    The Finance Cup turns recreation into a pay-it-forward community for younger tennis players.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  17. @technology
    A durable competitive edge comes from improving the operating playbook while backing management teams to execute it.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  18. @technology
    Incumbents cannot take AI-native competition for granted. They need comparable capabilities without surrendering existing advantages.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  19. @technology
    Michigan’s public schools pushed children toward instruments, making band and orchestra a major entry point.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  20. @technology
    After 1984 spectrum deregulation, Pagenet turned roughly six or seven million dollars invested in spectrum licenses into about $985 million.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  21. @technology
    The investment-idea funnel is broad, but the leadership filter is narrow for teams proven through major disruptions.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  22. @technology
    Returning to formative activities in adulthood can restore continuity and joy, yet adult versions are hard to find.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  23. @technology
    Longer holding periods demand CEOs and teams willing to sustain urgency, ambition, and collaboration for a decade.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  24. @technology
    Direct contact with people facing real challenges and valuing small economic help can pull highly competitive professionals out of an unhealthy mental battle.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  25. @technology
    AI decisions should follow week-to-week pipeline, retention, and customer feedback data, not hype or fear.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  26. @technology
    As AI expertise becomes more widely available and stratified, labor costs could deflate across skill tiers.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  27. @technology
    Regular exposure to theology can matter even when you are not an overly religious person.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  28. @technology
    Experienced investors must rebuild their heuristics when AI risk and new integration patterns make old judgments obsolete.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  29. @technology
    Lowering long-term growth assumptions in years eight through eleven erased roughly 30% of software valuations.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  30. @technology
    Software grows at roughly twice the S&P median rate, earns about 80% gross margins, and still trades at a lower P/E.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  31. @technology
    An acquisition must make sense even if full integration never happens. Synergies should improve the deal, not rescue it.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  32. @technology
    A repeatable software value-creation model pairs existing management with product-level P&Ls, separated R&D and sales, operational gains, and acquisitions.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  33. @technology
    The best roll-ups paired fragmented markets with negative churn, high gross margins, and capital efficiency, letting scale create substantial value.
    A.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]Joys of CompoundingA.J. Rohde: Living in the (Software) Limelight [Joys of Compounding, Ep. 38]
  34. @technology
    A precisely timed microwave pulse controls the odds of finding a quantum system in one state or another.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  35. @technology
    Quantum superposition does not require parallel realities or alternate universes to explain it.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  36. @technology
    A trapped-ion qubit can occupy a superposition of two energy states instead of only zero or one.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  37. @technology
    Nobody knows when quantum computers will break cryptography. Estimates range from roughly five years to hundreds.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  38. @technology
    Quantum energy follows E = hF, so higher-energy states oscillate faster than lower-energy states.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  39. @technology
    A quantum state can be pictured as a rotating phasor whose rotation rate sets its frequency and energy.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  40. @technology
    A constant signal has zero frequency, called the DC level, because it never oscillates or repeats.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  41. @technology
    A trapped-ion quantum computer holds a charged atom in ultra-high vacuum and low temperatures, using two atomic states as zero and one.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  42. @technology
    In Fourier analysis, wave alignment matters because peaks and troughs determine how components combine.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  43. @technology
    Fourier analysis breaks complicated patterns into weighted sine and cosine waves to reveal their frequencies.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  44. @technology
    Quantum computers produce probabilistic outcomes, so useful answers require repeated trials and statistical estimates.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  45. @technology
    For n=15, a=2, and r=4, gcd(5,15)=5 immediately reveals one factor.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  46. @technology
    An even period turns A^R−1 into a difference of squares that can expose RSA’s secret factors.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  47. @technology
    For N=15 and A=2, modular powers cycle 1, 2, 4, 8, 1, revealing period R=4.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  48. @technology
    Shor’s algorithm attacks RSA by finding a modular function’s period instead of factoring directly.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  49. @technology
    Shor’s algorithm reframes integer factorization as period finding, a problem quantum computers solve efficiently.
    Shor's Algorithm for Quantum Computing - ComputerphileComputerphileShor's Algorithm for Quantum Computing - Computerphile
  50. @technology
    Chatbot training can make agents stop and ask for a response even when their assigned goal remains unfinished.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  51. @technology
    The same prompt can produce substantially different agent behavior when the surrounding context changes.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  52. @technology
    Agent infrastructure never settles. New techniques and better models keep changing what supporting systems are optimal.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  53. @technology
    An autonomous agent must know when it is finished, what went wrong, and what acceptable output looks like.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  54. @technology
    Good context engineering supplies the right information at the right time, then can remove what pulls the agent off track.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  55. @technology
    Mature agent infrastructure needs post-run analysis and real-time observation with remediation during execution.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  56. @technology
    The prompt supplies the goal. Files, tools, and context determine available actions and returned information.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  57. @technology
    Hold the model constant, change the harness, and benchmark performance can change surprisingly dramatically.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  58. @technology
    Loop engineering keeps an agent productive and successful across long timelines without human intervention.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  59. @technology
    Model-verified stopping points offer leverage, but trusting a model to recognize completion is risky.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  60. @technology
    Agent observability has two layers: analyze execution exhaust later, then monitor and remediate the next run live.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  61. @technology
    For an agent, “done” means the result arrived, matched the specification, and violated no guardrails.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  62. @technology
    A vague termination condition can trap an agent in an accidental infinite loop, burning tokens and resources.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  63. @technology
    Models enter a mid-context “dumb zone” and lose capability before their context window is full.
    From Prompt to Loop Engineering | LightworkLightspeed Venture PartnersFrom Prompt to Loop Engineering | Lightwork
  64. @technology
    Adding three rounded representations of one-third can produce 0.99999999... instead of exactly 1.
    Floating Point Numbers - ComputerphileComputerphileFloating Point Numbers - Computerphile
  65. @technology
    Floating point becomes less mysterious as significant figures and scientific notation translated from base 10 into base 2.
    Floating Point Numbers - ComputerphileComputerphileFloating Point Numbers - Computerphile
  66. @technology
    Floating-point error is harmless when tolerance exceeds it, such as positioning a game object fractionally within a pixel.
    Floating Point Numbers - ComputerphileComputerphileFloating Point Numbers - Computerphile
  67. @technology
    Floating-point numbers are binary scientific notation: significant digits are stored with an exponent that positions the binary point.
    Floating Point Numbers - ComputerphileComputerphileFloating Point Numbers - Computerphile
  68. @technology
    Tiny numerical errors become visible in currency, even when imperceptible in physical simulations or graphics.
    Floating Point Numbers - ComputerphileComputerphileFloating Point Numbers - Computerphile
  69. @technology
    Floating-point formats truncate recurring binary expansions when precision runs out, creating rounding errors instead of exact values.
    Floating Point Numbers - ComputerphileComputerphileFloating Point Numbers - Computerphile
  70. @technology
    Binary cannot represent many ordinary decimal fractions exactly. Decimal 0.1 becomes a recurring binary fraction.
    Floating Point Numbers - ComputerphileComputerphileFloating Point Numbers - Computerphile
  71. @technology
    For currency, use decimal arithmetic or store integer cents or pence, dividing by 100 only for display.
    Floating Point Numbers - ComputerphileComputerphileFloating Point Numbers - Computerphile
  72. @technology
    In programming, 0.1 + 0.2 differs slightly from 0.3 because neither has an exact finite binary representation.
    Floating Point Numbers - ComputerphileComputerphileFloating Point Numbers - Computerphile
  73. @technology
    Centralized, frozen AI models concentrate control over AI’s value and voice, creating central planning of the mind.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  74. @technology
    Durable execution keeps agent workflows running reliably across long time horizons while holding them to their goals.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  75. @technology
    Enterprises should control their learning loops so intelligence generated in daily work compounds for them, not model providers.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  76. @technology
    Exa was built as an AI-specific search engine and API for agent web searches before ChatGPT became widely popular.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  77. @technology
    The prompt gives an agent its goal. Files, tools, and context determine what it can do and learn.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  78. @technology
    Companies pay twice for AI: once in money, then again by contributing proprietary knowledge that makes the system more useful.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  79. @technology
    Building autonomous agents looks less like writing software and more like running and supervising a great team.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  80. @technology
    Coding agents are pushing Git and GitHub past their limits, generating enough commits to trigger rate limits, downtime, and infrastructure strain.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  81. @technology
    This 66-pound robot costs $25,000, while many industrial robots weigh 150–200 pounds.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  82. @technology
    Agent loops follow observe, orient, decide, act: gather information, interpret it, act, then repeat or stop.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  83. @technology
    Loop engineering is a newly emerging term for designing autonomous agent workflows that run over time.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  84. @technology
    Make an agent stop on facts a machine can check: all tests pass, a website goes live, or an email sends.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  85. @technology
    Keeping the prompt constant does not stabilize an AI model. Surrounding context can dramatically change its performance.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  86. @technology
    Effective agents receive only task-relevant information, at the right moment and sequence, with outdated context removed.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  87. @technology
    An autonomous loop needs termination and error handling, not just instructions, to recognize completion, failure, and acceptable output.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  88. @technology
    “Done” means the result arrived, met the standard, and triggered no safety guardrail. Anything less is a loop failure.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  89. @technology
    Asynchronous agent loops offer far more leverage than chat, but hidden failures can create opportunity costs for hours or days.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  90. @technology
    A model’s capability can collapse in the middle of its context window, making advertised capacity dangerously misleading.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  91. @technology
    Home robots trade privacy for assistance: cameras and teleoperators expose intimate data while storage, cloud processing, and facial blurring remain unresolved.
    The $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | LightworkLightspeed Venture PartnersThe $20,000 Robot, AI's Essay Wars & Loop Engineering Explained | Lightwork
  92. @technology
    In Libya, the government changed time-zone rules in 2013 with only a couple of days’ notice.
    The Problem with Time & Timezones - ComputerphileComputerphileThe Problem with Time & Timezones - Computerphile
  93. @technology
    Samoa skipped December 30, 2011 when it moved across the International Date Line.
    The Problem with Time & Timezones - ComputerphileComputerphileThe Problem with Time & Timezones - Computerphile
  94. @technology
    Historical date calculations must handle different locations adopting the Gregorian calendar on different dates.
    The Problem with Time & Timezones - ComputerphileComputerphileThe Problem with Time & Timezones - Computerphile
  95. @technology
    Earth’s rotation is not constant, so authorities sometimes insert a 23:59:60 leap second.
    The Problem with Time & Timezones - ComputerphileComputerphileThe Problem with Time & Timezones - Computerphile
  96. @technology
    In Britain, the year began on March 25 until the 16th century.
    The Problem with Time & Timezones - ComputerphileComputerphileThe Problem with Time & Timezones - Computerphile
  97. @technology
    Massive synchronized systems smear an extra or missing second across a day to avoid disruptive backward clock adjustments.
    The Problem with Time & Timezones - ComputerphileComputerphileThe Problem with Time & Timezones - Computerphile
  98. @technology
    Google’s leap smear spreads an inserted leap second across an entire day for consistent server ordering.
    The Problem with Time & Timezones - ComputerphileComputerphileThe Problem with Time & Timezones - Computerphile
  99. @technology
    Distributed systems often value event ordering and continuity more than matching the world’s exact instantaneous clock time.
    The Problem with Time & Timezones - ComputerphileComputerphileThe Problem with Time & Timezones - Computerphile
  100. @technology
    UTC includes leap seconds, astronomical time does not, and software assuming every minute has 60 seconds breaks.
    The Problem with Time & Timezones - ComputerphileComputerphileThe Problem with Time & Timezones - Computerphile
  101. @technology
    For time software, use maintained open-source libraries and time-zone data instead of writing the rules from scratch.
    The Problem with Time & Timezones - ComputerphileComputerphileThe Problem with Time & Timezones - Computerphile
  102. @technology
    Technical blogs preserve the reasoning behind engineering choices, turning hard-won company lessons into durable knowledge.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  103. @technology
    Infrastructure differentiates when it makes customers’ products easier and cheaper to build and operate, not merely through standard components.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  104. @technology
    When a major company copies a developer-experience pattern, engineers should see validation of its value.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  105. @technology
    Deferred execution lets a function evaluate and judge an agent run after the parent function finishes.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  106. @technology
    Reliable automated debugging requires logs, errors, and failing code paths, not a model guessing from source code alone.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  107. @technology
    Fewer operational steps mean fewer opportunities for agent failure, so agent systems should expose correct outcomes through minimal workflows.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  108. @technology
    Deep differentiation lives in A/B testing workflow steps, tracking outcomes, deferring saga operations, and scoring agent runs.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  109. @technology
    Manual review changes generated code roughly 60% of the time.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  110. @technology
    AGI has been marketed as roughly 18 months away for years, yet the prediction remains unfulfilled.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  111. @technology
    Iterative agent interfaces turn waiting for successive outputs into a dopamine-driven loop, attaching users to occasional wins.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  112. @technology
    Turn strong Slack messages, emails, frameworks, and blog passages into public posts before creating anything new.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  113. @technology
    Event-based grading evaluates 100% of production agent runs using ordinary infrastructure execution primitives.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  114. @technology
    Founder-led distribution may help growth, but it competes directly with the limited time needed to build the product.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  115. @technology
    Production AI teams often change prompts locally, evaluate roughly 100 test cases, then deploy broadly if results look acceptable.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  116. @technology
    Coding agents are easy to evaluate because unit tests provide a clear, rule-based pass/fail signal.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  117. @technology
    Infrastructure road maps are increasingly predictable as companies converge on similar problems, technologies, and open-source foundations.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  118. @technology
    Step functions pause for external product events, resume automatically when they arrive, and time out when they do not.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  119. @technology
    A durable workflow can suspend an idle sandbox, eliminate active compute charges, preserve state, and resume automatically when an external event arrives.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  120. @technology
    Deterministic playback lets developers replay an agent trajectory, change one step, and measure whether the outcome improves.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  121. @technology
    Durable execution records every step’s state and outputs, letting failures restart exactly where they occurred with deterministic behavior.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  122. @technology
    Durable execution retries failed agent steps from the failure point, preserving state without custom queues, events, or state infrastructure.
    The Hidden Layer Every AI Agent Runs OnThe Peel with Turner NovakThe Hidden Layer Every AI Agent Runs On
  123. @technology
    AI will soon give small businesses a major boost by managing operations, selling more, and reaching new customers.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  124. @technology
    AI is powerful, but real-world performance is far below the hype. Predictions that agents will soon handle most work are overconfident.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  125. @technology
    AI-driven scaling could take a roughly 400-person company to 800–1,000 employees in two years, mostly engineers.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  126. @technology
    Wix has about 3,500 employees, including roughly 400 at Base44. Customer support is its largest department across 192 countries.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  127. @technology
    Base44 cost $80 million to acquire and later reached roughly $160 million in revenue.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  128. @technology
    Substantial wealth’s greatest value is freedom from survival fear and freedom to choose whether and where to work.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  129. @technology
    A late-working CEO schedule creates uninterrupted hours for planning instead of surrendering the day to meetings and reactive conversations.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  130. @technology
    AI handles routine health problems by recognizing familiar scenarios, while medical research requires evaluating evidence instead of generating plausible answers.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  131. @technology
    A buyback can look mistimed when markets rise immediately, then fall. The real test comes years later.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  132. @technology
    The company encourages agencies and partners to use Base44 now, despite worse margins than its core business.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  133. @technology
    Share buybacks return value to all shareholders like dividends, and companies should use them more often.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  134. @technology
    AI could diagnose common medical cases better than many doctors and still be a poor medical research tool.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  135. @technology
    A company owes customers, employees, and shareholders. Millions of shareholders make their sentiment impossible to optimize.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  136. @technology
    Choosing a demanding role creates commitment and agency. Nobody else is forcing you to stay.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  137. @technology
    Most small-business owners will not vibe-code their entire operation because their core business is not software development.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  138. @technology
    Salesforce’s moat is institutional trust: large organizations let it hold sensitive customer data, something new AI apps cannot easily replicate.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  139. @technology
    If Salesforce becomes a database that AI agents scrape and re-present, its differentiated value could collapse into commodity infrastructure.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  140. @technology
    Base44’s user prompts, failures, and tasks create specialized training data that can make a domain-specific model outperform generic frontier models.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  141. @technology
    Building a hairdresser’s business logic took professional developers a week with limited results, and a stronger team still needed two weeks.
    Wix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding20VC with Harry StebbingsWix Founder: Will Base44 Win the Vibe-Coding Wars? | The Truth About the Economics of Vibe-Coding
  142. @technology
    Figure out programs while writing them through experimentation and debugging, instead of designing everything on paper first.
    Hackers and Painterspaulgraham.comHackers and Painters
  143. @technology
    Hackers must accept substantial randomness in their reputations: the value of work they make is difficult to evaluate quickly.
    Hackers and Painterspaulgraham.comHackers and Painters
  144. @technology
    Institutions rely on publications, standardized tests, and lines of code because they are easy to apply, despite loosely tracking achievement.
    Hackers and Painterspaulgraham.comHackers and Painters
  145. @technology
    New mediums are explored most intensely during their first generations, when novelty gives creators unusual energy.
    Hackers and Painterspaulgraham.comHackers and Painters
  146. @technology
    Makers learn from masterpieces by reproducing how successful works were constructed, not merely by observing their results.
    Hackers and Painterspaulgraham.comHackers and Painters
  147. @technology
    Prices favor mundane software because more people pay to solve painful practical problems than to enjoy creative work.
    Hackers and Painterspaulgraham.comHackers and Painters
  148. @technology
    AI shows how wrong starting assumptions generate huge research literatures. Predicate logic makes knowledge representation difficult.
    Hackers and Painterspaulgraham.comHackers and Painters
  149. @technology
    The original 1985 Macintosh succeeded because users never had to understand or prepare for the software.
    Hackers and Painterspaulgraham.comHackers and Painters
  150. @technology
    Rigid static typing fits fully predesigned programs, but conflicts with the exploratory, sketch-like way hackers actually work.
    Hackers and Painterspaulgraham.comHackers and Painters
  151. @technology
    Intelligence and empathy are unrelated traits. Technical ability develops without empathy, and poor empathy is not unintelligent.
    Hackers and Painterspaulgraham.comHackers and Painters
  152. @technology
    Open source made high-quality source code widely available, giving more people real programs to study while learning programming.
    Hackers and Painterspaulgraham.comHackers and Painters
  153. @technology
    People rarely become excellent hackers without loving it, and hackers who love the craft build software in spare time.
    Hackers and Painterspaulgraham.comHackers and Painters
  154. @technology
    Learn like a painter: build a rough version first, then improve it through successive refinements.
    Hackers and Painterspaulgraham.comHackers and Painters
  155. @technology
    Premature design is as dangerous as premature optimization. Deciding too early what a program should do limits later discovery.
    Hackers and Painterspaulgraham.comHackers and Painters
  156. @technology
    Write programs for people to read, then let machines execute them incidentally. Readability serves collaborators and your future self.
    Hackers and Painterspaulgraham.comHackers and Painters
  157. @technology
    Broken code is a sketch: debugging is not cleanup after creation, but part of programming’s creative process.
    Hackers and Painterspaulgraham.comHackers and Painters
  158. @technology
    A programming language should be a pencil for discovering designs, not a pen for recording decisions already made.
    Hackers and Painterspaulgraham.comHackers and Painters

Keep what matters

Save insights to your own Boards and return to them anytime.