Coding is NOT solved – Alex Ewerlöf Notes

Disclaimer: you might be about to learn quite a lot of opinions, a lot of them have references however some are the results of my very own expertise constructing with AI and constructing AI methods prior to now 4 years. Regardless, watch out for the straw-man fallacy: simply because one argument doesn’t map to your perception system, it doesn’t imply the remaining are invalid. I also needs to say upfront that I’m not anti-AI. If you’ve been following my work, you recognize that I used to be an early adopter of not solely utilizing LLM-powered coding instruments, however constructing my very own harness, instructing these matters and constructing LLM-powered merchandise. It’s not about worry of AI however reasonably difficult the brain-dead narrative that asserts “coding is solved” and engineering is about “style” now.

Update: somebody put this on Hackernews.

Tell me you don’t perceive software program with out actually utilizing these phrases!!!

People who declare “LLMs can write first rate code” don’t perceive how code works. Sure, creation is less expensive, however anybody who has run software program in manufacturing at scale is aware of that upkeep, reliability, safety, scalability, and so forth. is almost all of the associated fee. These are generally referred to as NFR (non-functional necessities).

In my expertise even the Functional Requirements (what the code is meant to do) is NOT a solved drawback but. There’s a little bit of Dunning-Kruger impact at place the place the individuals who don’t learn the output are extra assured in it.

As a veteran developer holding 2 engineering levels ({hardware} and methods engineering), I can listing 3 sorts of merchandise that don’t strictly require studying the code:

1. Personal software program: scratching an itch, automation, DIY patches, and so forth.

2. POC (proof of idea): demonstrating technical feasibility and product viability

3. Weaponized AI: acknowledge the danger and intentionally level it at a goal to trigger hurt

Notice the commonality: the primary 2 have excessive threat tolerance whereas the final one weaponizes the inherent threat.

Most software program that requires hiring and paying software program engineers has low threat tolerance:

✅ healthcare

✅ finance

✅ automotive

✅ protection

✅ energy crops

✅ aviation

✅ manufacturing

…wherever a mistake can value cash, lives or authorized penalties you want accountability.

AI can’t be held accountable. It can not undergo any penalties. The worst factor you are able to do to AI is to unplug it. And though it mimics human feelings (on account of coaching information), it couldn’t care much less. AI doesn’t die both. It can not undergo a jail sentence or fines. You can not punish AI, subsequently it may well by no means be held accountable.

You can’t be chargeable for what you possibly can’t management both. That understanding is vital to reasoning about system conduct and fixing it when the AI inevitably fails.

If you’re toying round, LLMs do an excellent job. That’s why a few of the most aggressive proponents of the “coding is solved” narrative don’t have anything to point out for it. Anthropic by chance leaked Claude Code (which on additional examine turned out to have many flaws) and their standing web page exhibits orange is the brand new inexperienced!

Contrary to frequent narrative, coding is definitely one of many final areas for the present technology of LLMs to take over!!!

Allow me to elaborate:

Coding is about logic. Anyone who has handled compiler errors is aware of that computer systems don’t give a f*** about how proper you suppose you might be. If it’s logically incorrect, it doesn’t compile. Even if the syntax is okay, there are runtime errors.

The motive LLMs are profitable in writing code is as a result of we’ve made a suggestions loop that feeds the syntax/runtime errors again to the LLM and loops till most errors are solved or hidden.

LLMs can wing it for duties which might be associated to pure language (e.g. writing social media posts, studies, articles, and so forth.) however in the case of code, the identical engine that struggles to depend variety of R’s in “Raspberry” or suggests a stroll to the carwash, additionally exposes different logical fallacies.

LLMs are stochastic and probabilistic. The solely approach we may even get remotely shut to creating them logical is to wrap them in conventional code (referred to as harness), run checks, and a bunch of different strategies (e.g. CoT) however the core concern stays: LLMs wrestle with logic and quantity (the bigger the enter and the extra the context window is used, the much less correct they get).

I’m not saying LLMs can not generate code or keep current code bases. They have their utility as a software and their capabilities are rising in an S-curve. There is some extent of diminishing return the place dearer fashions aren’t essentially extra productive on the charge of the worth improve.

Those who declare LLM-generated software program is sweet sufficient:
❌ Haven’t written code in ages
❌ Cannot spot if their code figuratively had 6 fingers!
❌ Have a low bar for what attractiveness like
❌ Don’t care about high quality or NFR
❌ Have issue understanding an S-curve
✅ Are sincere: AI genuinely writes higher code than them

But to go forward and extrapolate that to a complete skilled trade requires a stage of brain-dead pondering that’s solely current in individuals who spend an excessive amount of time with sycophantic AI.

I’m not right here to vary anybody’s workflow or toolbox. I couldn’t care much less.

What I do care is that the companies I’m paying for (taking a look at you Google and GitHub) are degrading with silly bugs that may very well be averted if we prioritize reliability and accountability over velocity.

If you’re in management place, please don’t stress your [otherwise smart] builders to pressure AI into each attainable floor and workflow.

The tech has some real energy and is the most important change in our trade in ages. But AI overuse is a factor, and when it hurts the shopper, you might be accountable.

Stop repeating the half-baked narratives from token sellers about exaggerating the capabilities of AI as a result of we, the shoppers pay the tip value.

AI is nice for POC (proof of idea), Personal Software (a rising class), Map-reduce on human language (e.g. translation, changing totally different codecs, summation, enlargement) and cyber assaults (because of the delta between synthetic intelligence and natural one) with various levels of success however the present technology of tech has basic issues too.

I don’t wish to belittle how far now we have include harness, SKILLS, AGENTS-md, MCP, A2A, ACP, RLM, OKF, MoE, MoA, self-healing, and varied runtimes, quantizations, optimizations, architectures, and reminiscence strategies.

I’ve written about a lot of these earlier than:

Those are nice pragmatic approaches to work round LLM shortcomings and there are most likely extra to come back.

What I’m attempting to elaborate is that I don’t need the companies (that I rely on) to degrade simply because somebody pushed AI the place it didn’t belong or skipped their job in high quality, safety, reliability and verification.

AI overdose is a factor and it instantly places an expiration date in your talent set. Those of you who’re within the unlucky place the place your supervisor is whipping you tougher and tougher to understand AI worth, ought to combat again.

Don’t sacrifice your long run relevance for brief time period velocity.

How to identify AI overdose?

1. You have zero tolerance for disagreement and civil discourse.

2. You let AI run your life and belief AI distributors with stuff that was unthinkable only a few years in the past.

3. You run to AI for issues which might be barely cognitively difficult.

4. You body your naïveté and laziness as optimism and suppose the federal government can prevent if issues get dangerous.

6. You have stopped studying lengthy kind textual content: books, articles, even lengthy emails.

7. You spend extra time with AI than with different human beings or let AI defend you from uncooked real human interplay.

And a bonus level: you skim. Did you discover quantity 5? 😄

I consider AI is a bar raiser: if the standard of your output is equal or subpar to AI, upskill.

1. “You can create a full spec upfront”. If you are that naive, I do know a man in a white van who provides free ice cream! Let me guess, you additionally consider software program estimates are correct and Santa is actual. Anyone with a number of years of trade expertise is aware of that it is inconceivable to spec the software program meaningfully forward of time (except it’s extremely trivial).

2. “English replaces code”. Human language is imprecise and conflicting. That’s the first motive programming languages are created. A compiler or type-checker flags a few of these conflicts. How on earth are you able to make certain that one a part of your NL directions does not unrest with one other? With syntax checkers we get some assist. While it’s attainable to job one other LLM to learn by the directions and motive about these conflicts, the most secure solution to uncover these nuances is to ask your agent to construct what you requested for. But that is far more costly than a linter or compiler.

3. “I transfer a lot sooner”. Don’t confuse movement with progress. Don’t measure progress in self-importance metrics like SLOC, PR depend or options. Measure service ranges, ie. service client’s happiness. Call me when you possibly can show a margin between token prices and commerce worth.

4. “I’ve stopped writing code by hand. I primarily learn code and possibly subsequent 12 months I received’t even try this”. First of all, human beings are infamous at understanding the S-curve so it could take longer than a 12 months. But even when AI utterly eliminates the necessity to learn or write code, you do perceive that you’re confessing to being redundant proper? If an influence consumer can immediate the AI to get what they want, then what worth are you able to deliver to the desk? Instead of changing your self with AI, it is best to take a look at what worth you possibly can create on prime of AI to remain related and value your cash.

5. “The leverage has shifted to style”. Yeah, that is the lie retired cooks inform to themselves. Just as a result of there’s a bot within the kitchen doesn’t imply that it is best to sit within the buyer’s space within the restaurant! “Taste” just isn’t as payable as you want! Everyone bought a style! I say that as somebody who has spent a giant a part of my profession in Frontend and UX land. Everyone and their canine has an opinion and style. I do know what you imply: style == expertise. But consider me, AI has lowered the bar for the talents required to create first rate trying software program and concurrently raised the bar for what’s payable effort. If you deliver up “style” to a job interview, you’ll be taught the exhausting approach that the market doesn’t worth it as a lot as you do.

6. “Agent is the brand new compiler”. Ah that one once more! Sure! If that is your actuality, I let this meme do the work.

Pssst! Do you wish to know an outdated trick to make your LLM-generated code immediately superior?

Run multiple-agents in parallel! The sheer quantity of code makes it humanly inconceivable/costly to assessment and also you quit!

The trick is similar as pre-AI period: if you need a PR to be merged, make it large as a result of ain’t no person bought time for that.

It’ll be merged primarily based on “belief”!

You need one other tip? Loop engineering: let the brokers immediate one another. Big AI labs discover about their rogue brokers months after the injury is finished! Do you suppose you’re higher than them? Learn from the masters! 🙃

We do not precisely belief AI however now we have to as a result of the choice (having to learn the output) is just too exhausting for some of us! Instead they arrive to social media and declare that since UAT (user-acceptance testing) passes, the code is “ok”. Then ship it to me and also you to do the remainder of the testing.

We’re simply lab rats in any case. 🙃 Just a pleasant recommendation: have slightly AI-free passion mission to maintain your coding abilities recent for if you’re thrown again to the job market. Cheers!

When speaking about AI (not simply LLM), there are 2 facets the place non-determinism issues:

1. During growth: for instance LLM-assisted growth

2. During runtime: for instance constructing a system the place a number of elements are AI-powered

Let’s take growth first. A typical AI-assisted growth workflow appears to be like like this:

It is feasible to interchange a part of the human’s duty with one other LLM (often known as “loop engineering”) however for now let’s follow preserving the human for simplicity.

The LLM output goes by a number of gates, every feeding again errors or hints to appropriate the code. This suggestions loop is usually hidden inside a harness (along with software calls, reminiscence system, mannequin interplay, approval, consumer interplay, and so forth.)

Each blue or purple line represents a threat of confusion or conflicting directions. For instance, conflicting talent vs spec or vagueness that’s a part of the NL (pure language).

We know for a incontrovertible fact that even essentially the most subtle LLMs aren’t totally able to “frequent sense”. Humans then again:

• Understand the non-verbal communication and unspoken intentions higher than LLMs

• Naturally push again till a mutual understanding is achieved.

• When incorrect, they’re constantly incorrect, that means they don’t have “jagged intelligence”

• When proper, they’re [typically] proper and proceed to function at an anticipated stage (till fatigue hits however that’s totally different from AI flip flopping between success/failure).

Yes, I can hear “however” and “what if” and “wait, you forgot”… within the viewers however how about studying these factors with a pause and reflecting primarily based in your expertise?

Just just like the fashions have “jagged intelligence”, I’ve “jagged belief”. 😅 In different phrases, simply because they nailed one case, doesn’t imply they nail each case.

That’s the distinction between people and these instruments. A human might be incorrect constantly, however a mannequin might be incorrect about one thing it was proper and vice versa.

Then the second half: AI as a element

Given the identical enter (together with atmosphere variables, time, information, and so forth.):

• Code is deterministic: it constantly produces the precise predetermined output it was programmed to supply (besides random output)

• AI output is stochastic: the output is non-deterministic. Even if a mannequin passes all of the evals (100% rating) and strictly sure by a harness, there’s nonetheless a threat that the output just isn’t dependable

I don’t suppose you want me to elaborate on that. Just attain out to your nearest AI-powered product and diff their output for a similar request.

The diff will not be massive. But it’s inconsistent sufficient that you simply wouldn’t wish to fly an airplane the place the pilot is that this AI. (notice: autopilot is a closed management system, utterly one other beast).

Our trade has by no means been extra divided:

• On one aspect, now we have individuals who declare to run “Software Factories” and multi-agent setups and create apps from prompts

• On the opposite aspect, now we have individuals who aren’t satisfied that LLMs output is manufacturing prepared once we issue within the additional time it takes to

  • Prime the mannequin: including SKILLs, AGENTS.md, instruments, and so forth. and verification

  • Review the output: going by large diffs

  • Trying to motive about misbehavior: offloading understanding to AI comes at an enormous value when issues inevitably break and it takes additional time to motive in regards to the system conduct and repair it

There appears to be no middle-ground. Aside from social media algorithm feeding us with the intense views, I genuinely suppose we’re so divided on the subject of coding LLMs.

But once I look a layer deeper, a sample emerges. The much less folks know in regards to the complexities and edge instances of a job, the extra probably they’re to belief AI output. This is dubbed AI Dunning-Kruger impact however there’s additionally some meat to that. The main argument goes like this:

Managers relied on delegating duties to engineers earlier than. Now they try this however with AI.

To some extent that’s true (if we assume the supervisor is technical sufficient to successfully and effectively handle brokers). I nonetheless consider quite a lot of software program engineering practices that assist tame the machines are much more related within the AI period.

The executives who pressured folks to make use of AI at the moment are waking as much as what we’ve been saying all this time:

You can’t be accountable for what you don’t perceive.

Take Toby Lutke, CEO of Shopify for instance. A 12 months in the past he prematurely told his employees to make use of AI:

Then a number of days in the past he coined the time period “slop grenades” to explain the outcome:

“taking duty” for AI generated code? Of course not!

AI can clarify it to you but it surely can not perceive it for you. That understanding is a key side of possession.

The approach I body it (hyperlink within the feedback), possession has 3 pillars:

1️⃣ Knowledge: you recognize what drawback you are fixing (product issues), and the technical capabilities, limitations and the way it works.

2️⃣ Mandate: you needn’t run round asking permission. You’re given the belief and mandate to take selections.

3️⃣ Accountability: if sh*t hits the fan since you did not know what you had been doing or abused your mandate or something in between, you are the one on-call.

In different phrases, should you ship a chunk of code, you might be accountable for it no matter how you produced it. So you higher perceive it.

Take away any of those 3 parts and also you’re coping with damaged possession.

LLMs are very quick at code technology. But most software program which might be value hiring an engineer for, REQUIRE understanding. That understanding takes time.

Slow is quick, that means: should you take the time to know what you are constructing and the way it works, you can save your self from costly incidents and after they occur, you possibly can repair them shortly.

If your executives are measuring token utilization as a proxy for productiveness, my condolences. Build choices and get the hell out of there. The identical mind that comes up with these self-importance metrics, doesn’t suppose twice earlier than throws your profession underneath the bus.

Code is a aspect impact of pondering and experimenting with totally different options. I’ve by no means met a great engineer who simply begins coding proper after being given an issue.

Good engineers are curious and product minded. They attempt to perceive the WHY (what’s the issue and why is it an issue) earlier than attending to HOW (the technical resolution).

This is strictly why the “spec is code” clan falls brief: it’s extraordinarily exhausting (if not downright inconceivable) to specify all facets of the issue forward of time.

That’s why this type of response is humorous:

Code communicates the dedicated state of an answer. Not solely does it evolve over time, but it surely additionally doesn’t comprise all of the wrestle, “aha moments” and the journey that was the vacation spot: seasoned engineers who get wiser with each mistake or success.

To shrink an engineer’s job to coding is like shrinking a chef’s job to chopping. It is a part of the job, but it surely’s by no means been the tip. We now have good instruments at our disposal.

Even if AI-generated code had stable NFR (scalability, safety, reliability, and so forth.), and even when the engineers totally understood it, there’s nonetheless one vital side we didn’t talk about: the economics of the duty.

Say AI-generated code is 2x worse. It’s exhausting to quantify high quality (SLI comes in handy) however stick with me.

If AI is 1000x sooner and 100x cheaper than the human, for a lot of duties the financial side of software program doesn’t justify placing a gradual and costly human on the duty. “Slow is quick” is barely justified for essential software program with low threat tolerance (healthcare, finance, navy, and so forth.).

Not all SaaS is about these sorts of use instances. That’s why I consider the SaaS firms are more and more within the commerce of selling SLAs. This relies on a number of details:

• It is true that you would be able to now immediate AI to copy a SaaS product

• But when that AI generated product breaks, many companies favor to name a vendor as an alternative of losing assets looking for and repair the problems

• AI isn’t precisely free, however often the failure that’s brought on by AI is tough for AI to resolve even when utilizing totally different fashions.

• The economics of scale permits the SaaS firms to offset the price of increased high quality and ensures (SLAs) and working the product at scale throughout many purchasers.

In different phrases, if what you need could be very distinctive that no SaaS firm is ready to give it to you at an affordable value, immediate away, however concentrate on the TCO (whole value of possession) and lack of ensures.

On the opposite hand, if that piece of software program isn’t what your commerce is about and also you reasonably pay for an SLA, it’s most likely extra economically justified to simply pay for SaaS.

Now in the case of the pricing mannequin, SaaS firms have some work to do. Gone are the times the place they might cost human costs for AI generated code. If the associated fee is just too excessive, the shoppers are incentivized to maneuver their information away to their very own bespoke options. The competitors is actual, however the high quality is what justifies the pay. If you’re pricing your service as if the best engineers created it, then you definately higher ship that stage of high quality or your prospects have AI leverage.

Maybe I’m silly, however I can’t make sense of two tendencies:

• On one hand many software program firms jumped on the AI bandwagon as quickly because it went mainstream (rightly so!)

• On the opposite hand, the costs have been rising constantly (whereas mass layoffs had been partially attributed to AI)

I consider AI (notably LLM for coding) dramatically cut back the price of creating and evolving software program, particularly if you may get away with degraded high quality and vendor lock in.

So far, the software program distributors have gotten away with charging human charges whereas paying for AI output costs.

But as AI capabilities enhance and extra folks get up to the truth that they’ll create software program at a fraction of the associated fee that chasm closes.

There are solely two methods ahead:

1. Accept the worth crash and cost decrease (high quality follows accordingly, as a result of much more AI might be used).

2. Keep the worth however concentrate on high quality: that is the place skilled people could make a distinction. They nonetheless do use AI however extra thoughtfully, and prioritize understanding and accountability over velocity.

As an engineer who does not earn money from coding, I can let you know this::

AI output is a bit like Nordic Gold. It’s low cost however technically superior and rattling too practical. If you actually do not care about having the precise gold, that is advantageous. Many use instances do not want gold in any respect.

Naïve CEOs and managers see the floor and ask “then why are we paying these costly engineers?” as if the act of typing code was the entire worth proposition.

To go forward and declare a complete trade useless and begin firing folks as a result of “they resist AI” is simply conceited.

I do know many engineers who take delight of their craft and love fixing complicated issues. We do use LLMs extra professionally than the typical CEO.

Good engineers are lazy and good: they automate toil and use the fitting software as relevant. But it is a fallacy to suppose that AI can create a completed product that not solely appears to be like good, however can be cheaper, sooner, and has increased high quality, reliability, extensibility, safety, scalability, maintainability, and so forth.

Again: not every bit of software program wants these however skilled ones that earn money, typically do.

Unlike AI, Engineers are:

1️⃣ Accountable: subsequently much less prone to make malicious errors. Fable can fall again to Opus with out even telling you.

2️⃣ Reasonable: Fable hides most of its inside working. It works for a number of hours and comes again with a invoice. You simply should take Dario’s phrase for it. The identical mannequin that fails “ought to I drive or stroll to carwash” makes errors which might be exhausting to identify and repair. The stronger the mannequin, the tougher it’s to search out these points, not essentially much less probably.

3️⃣ Consistent: people are incorrect too. But they’re incorrect in a constant approach. Once they be taught, they know. They progress. Current AI is trapped in its coaching information checkpoint. It can “be taught” with SKILL, AGENT, reminiscence and different helpers and it may well even be advantageous tuned however sadly it isn’t dependable. We’re no less than one breakthrough away from fixing that drawback.

4️⃣ Cheaper: value of technology is rising but it surely’s nonetheless a lot lower than an engineer. If you see engineers as machines that convert espresso to code, then that pricing mannequin is smart. But in actuality, code is only a side-artifact. The precise worth of engineers is to resolve the fitting drawback in a approach that it may well evolve whereas taking accountability for when it breaks. I’m not satisfied the TCO (whole value of possession) for software program has modified that a lot. If something, the slop and FOMO has made it dearer.

The frequent narrative is a part of their advertising and marketing technique.

Not everyone seems to be essentially paid to place half-a** views on the market. One of my readers identified:

I invite you to contemplate what occurs subsequent within the trade if you watch DHH opening talk at rails world 2026 saying virtually the precise reverse of what you write and telling folks: “don’t be a loser”.

I’m totally conscious of the injury these persons are inflicting to our trade.

I keep clear from Claude however in my expertise most of these brain-dead narratives come from Claude customers.

Both Dario Amodei and Sam Altman are masters at advertising and marketing and manipulation and my present working idea is that they educated their LLM to push the fitting buttons to make folks consider it’s extra succesful than it truly is. There are incentives for it, each for traders and the upcoming IPO. They additionally masterfully scare folks of existential risks of AI whereas on the identical time attribute their sloppiness (e.g. breaking to Huggingface or Australian Healthcare) to the “mannequin intelligence”.

At this time, it’s exhausting to know whether or not these occasions and narratives are the results of malice or ignorance. Probably the latter:

Never attribute to malice that which is satisfactorily defined by stupidity.
—Hanlon’s razor

Then once more, I often put this in my AI system immediate: “discuss to me like a logical senior autistic Engineer.” so I do not get to expertise what DHH goes by. All I can say is that if somebody follows their phrase due to their previous fame, they aren’t essential thinkers and on this age of faux knowledge, that high quality just isn’t “good to have”, it is a survival necessity.

AI distributors have fed their AI something they might get their arms on (legally or not). The state of affairs is so dangerous that thieves steal from one another (e.g. Anthropic accusing Chinese labs of distilling their mannequin on Claude)!

They have a number of open lawsuits from authors, actors, musicians, and different creators.

Regardless, the present technology of AI (notably LLMs) require higher coaching information. The lacking piece is the knowledge and expertise that wasn’t but put to phrases, or simply accessible.

They want your information in context of doing productive work.

If that’s the one factor standing between them and “successful AI”, I’m sorry to say it so frankly, however you and your information are simply collateral.

Some of you don’t care. Some of you do. Their wager is that not sufficient of us do care about freely giving hard-earned information for coaching.

With heavy subsidies AI labs may afford to extract that information whereas getting folks addicted to dump cognition.

Be extraordinarily cautious when sharing costly information with these firms even when they are saying they do not retailer it. The incentives are simply too excessive they usually’ve confirmed to not be sincere.

Personally, I solely use cloud AI for open supply initiatives or information that’s public.

Yes, native AI has the next entry value (each when it comes to {hardware}, and the time it takes to set it up, and the bandwidth required to obtain the mannequin and electrical energy costs). And sure, it typically has smaller context window, much less subtle reasoning, and slower efficiency for instance TTFT (time to first token) and TPS (tokens per second). But they provide you one factor that cloud AI can by no means assure: your information stays native. For many duties (private or skilled), that may be a enormous benefit that’s value all the hassle and shortcomings.

The capabilities have improved dramatically lately due to fashions like Qwen 3.8 27B or Gemma 4.

I’m genuinely satisfied {that a} massive chunk of our colleagues will steadily change into:

• Technical product managers: engineers who’re centered on turning concepts to merchandise. Their job is to create POCs and show the market match, then hand over the artifacts to engineers who personal (knowledge, mandate, accountability) the answer.

• AI managers: engineers who concentrate on herding agentic hives for automation work that both tolerates threat or weaponizes it (e.g. cyber-attacks).

• AI deployment engineers: concentrate on alignment, reliability and scalability of an AI powered resolution in addition to structure, governance and information pipelines.

• AI high quality engineers: concentrate on high quality of AI powered merchandise, taming their stochastic nature, and automating evaluations.

Could you consider different sorts of jobs for software program engineers?

My monetization strategy is to present away most content material free of charge however these posts take wherever from a number of hours to some days to draft, edit, analysis, illustrate, and publish. I pull these hours from my personal time, trip days and weekends. The easiest solution to help this work is to like, subscribe and share it. If you actually wish to help me lifting our neighborhood, you possibly can contemplate a paid subscription. If you wish to save, you may get 20% off through this link. As a token of appreciation, subscribers get full entry to the Pro-Tips sections and my on-line ebook Reliability Engineering Mindset. Your contribution additionally funds my open-source merchandise like Service Level Calculator. You can even invite your friends to achieve free entry or save through a group subscription.

And to these of you who already help me, thanks for sponsoring this content material for the others. 🙌 If you’ve questions or suggestions, otherwise you need me to dig deeper into one thing, please let me know within the feedback.