What we learn building software and AI.

What breaks first when you scale a product too fast
Scaling problems do not arrive uniformly. The same few things break first on almost every product that grows faster than its infrastructure.
Read
How to read a build cost estimate you did not write
A cost estimate is a set of assumptions wrapped in a number, and the number is usually the least useful part of it.

When an agent should ask for help instead of proceeding
An agent that proceeds when uncertain is dangerous. An agent that asks too often is useless. The design question is how to find the right threshold between them.

How to run a technical steering committee
A technical steering committee only helps if it makes decisions rather than just reviews them, and most of them never get to that point.

Why context windows change how you structure a task for an agent
Every token the agent reads is a token that gets used before you get an answer, so the structure of a task determines what the agent still knows when it starts writing.

How to define done on an AI build
When AI writes the code, done needs a sharper definition than it did before, because the gap between plausible-looking output and working software is wider.

How to negotiate timeline when scope cannot shrink
When a deadline and a scope are both fixed, the argument only sounds like it is about scope, because the real choices are quality, cost, and how much work runs at the same time.

What good error messages do that most don't
Most error messages tell a user that something broke and stop there, which is the moment a small problem turns into a support ticket or a lost user.

How to decide when to sunset a product
A product that costs more to keep running than it earns or gets used is an easy decision, but shutting it down well takes a full plan. A notice that it is going away is too little.

Why integration tests catch what unit tests miss
A unit test proves a function is correct on its own. It cannot prove two correct functions agree on what they are passing each other, and that gap is where a specific type of bug keeps appearing.

How to tell a real blocker from a preference
If the work can proceed a different way, it is not blocked, and calling it that teaches a team to stop believing the word.

How to run a bug bash before launch
A group testing session that only produces a list of typos was set up wrong. The fix is who you invite, how you plan the session, and how you sort what it finds.

Why most technical debt is a decision, not an accident
Technical debt usually gets blamed on time and sloppiness, but in the work we do it almost always traces back to one moment where a team chose speed over the fuller answer and never wrote that choice down.

How to decide if you need a platform team
A platform team is worth its cost once shared infrastructure work starts taking hours from every product team at once, and not before.

What to check before you sign a software development contract
Four clauses decide whether a software contract protects you or just the vendor, and most founders only read them after something has already gone wrong.

How to write a postmortem that changes behaviour
A postmortem with no owner and no date on its action items is a document that reads well and changes nothing.
From AI News
Short briefs on AI and software engineering, published as things happen.
Bitwarden opens Agent Access, a protocol for giving agents credentials
SageMaker now watches inference endpoints and flags cheaper setups by default
GitLab Duo Agent Platform credits now stretch up to twice as far
Linger puts a local personal agent on your own PC
Onlook rebuilt its design tool as an agent that edits real code
Guides
In-depth, practical guides from a team that builds and releases software.
How to choose a software development partner
Most of the cost of building software comes from choosing who builds it with you, and from getting that decision right before the first line of code is written.
How to build an AI product that reaches production
The hard part of an AI product is everything between a demo that impresses and a product people trust every day. This guide is about that work.
How to launch a product: from idea to MVP
An MVP is not a smaller version of the product you imagine. It is the fastest honest test of whether the product is worth building at all. Get that framing right and everything after it (scope, stack, team, launch) gets simpler and cheaper.
Technical due diligence: a complete guide for investors
Technical due diligence is a focused review of a company's software and the people who build it, run before an investment or acquisition, to check whether the technology can support the plan being paid for. It covers code quality, architecture and scalability, the engineering team, security and compliance, and technical debt, and it ends in a priced list of risks the deal team can act on. This guide explains what the review is, how it runs on a deal timeline, what it costs, and which findings change the price or the terms.
AI startup due diligence: how investors check what is real
AI startup due diligence is the work of checking that an AI company's product does what the pitch says, that the code behind it can be maintained and sold, and that the margin stays positive after the model-provider invoices. It differs from ordinary technical diligence in three places: the claim can be tested live instead of inspected, the codebase was probably written by a model, and the cost of goods is a variable bill from a vendor who can change the price. This guide covers all three for VC associates and principals, private equity deal teams, and angels in AI deals.
Angel investor due diligence on technology: judging a startup's software
Angel investor due diligence on technology is the part of the checklist that the published angel checklists leave out. The Angel Capital Association's 2007 best-practice guide has no section on software or code, and the UK Business Angels Association's 2020 checklist asks eight technical questions, all of them about positioning and intellectual property. This guide covers that missing part for angels, angel group members and syndicate leads, most of whom do not write code. It covers what to check in a screen share, what an MVP claim should mean, how to see a single-developer risk, how to judge a technical founder, and when to pay a firm to look properly.
Technology after the deal: the engineering work that follows an investment
Technology after the deal is the engineering work that starts when the investment closes: the technology workstream in the 100-day plan, merging or separating codebases, sizing technical debt across a portfolio, choosing who leads engineering, and reporting progress to the fund in measures a board can read. This guide is for private equity operating partners, venture platform leads and portfolio company board members. It covers what to do in each phase, what to check, the mistakes that repeat, and the timelines and costs that third parties have measured.
How to prepare for technical due diligence: a founder's guide
Preparing for technical due diligence means assembling, before an investor asks, the evidence that your software does what your deck says, that your team can keep building it, and that nothing in the code, the licences or the state of your security will surprise anyone after the investment is made. This guide is written for founders and CTOs raising a round or selling the company. It covers what a technical review asks for, what to put in the engineering data room, how to present your architecture and roadmap, how the review changes from seed to Series B and at a sale, what changes when the code was written by AI, and a day-by-day checklist for the two weeks before the review starts.
How to scale your engineering team
Scaling an engineering team is not about hiring faster. It is about adding the right capacity at the right time, in a way that makes the team quicker rather than slower. Most teams get the timing and the method wrong, and pay for it in speed.
Custom software: build vs buy, done right
Most companies do not need custom software for everything. They need it for the few things that make them different, and they need to buy the rest. This guide is about telling those apart, then building the custom part in a way you will not regret.
What software costs and how to budget it
The price of software is not one number. It is the result of choices you make about scope, seniority, and how right the product has to be. This guide explains what actually drives cost so you can build a budget that stays accurate, instead of relying on a quote that does not.
From AI-generated code to production software
AI can write most of a codebase in an afternoon. It cannot tell you whether that codebase is safe to run. This guide is about the difference between the two, and how to deal with it.
How to reduce Claude Code token usage
To reduce Claude Code token usage, make the conversation shorter and keep repeated text at the cached price. Claude Code sends the whole conversation with every request, so usage grows with the length of the session, by Anthropic's documentation read on 4 October 2026. The changes, in the order we recommend: measure with /usage and /context, run /clear between unrelated tasks, keep CLAUDE.md under 200 lines, filter test and log output, use Sonnet for most work, and write requests that name the file. Anthropic states an average cost of $13 per developer per active day. This guide explains each change and links to a page on each.
The Claude Code prompt cache: how it works and what resets it
The Claude Code prompt cache is a store of request text that the service running the model has already processed. Claude Code sends the whole conversation with every request, and the cache lets the unchanged start of that request be billed as a cache read, which Anthropic's pricing page, read on 4 October 2026, lists at 0.1 times the base input price on most models. The match is exact, so a change early in the request recomputes everything after it. Anthropic's documentation lists nine actions that can reset the cache, such as switching models, and eight that keep it, such as editing files. A stored copy lasts five minutes or one hour without use. The /usage command shows how much of your input came from the cache.
Claude Code costs for teams: what it costs, how to see it, and where to cap it
Claude Code costs a team, by Anthropic's own figures read on 4 October 2026, an average of $13 per developer per active day and $150 to $250 per developer per month, with costs below $30 per active day for 90 percent of users, and Anthropic says the figure varies widely with model choice, codebase size and habits. How a company pays depends on its setup: a Claude for Teams or Enterprise plan with a seat allowance, the Claude Console billed per token, or a cloud provider billed to the cloud account. This guide is for the manager or finance lead who approves the spend. It gives the figures, the setups, the screen and the cap for each, the limit messages, the settings an organisation can enforce, and a first-month plan.
The cost of agents and unattended runs in Claude Code
Every extra agent in Claude Code is an extra context window, and every context window sends its own billed requests. A subagent, an agent teammate, an agent in a dynamic workflow, a scheduled task firing while you are away, and a `claude -p` run started by a script all add requests that the main conversation never shows you. Anthropic's documentation, read on 4 October 2026, states the cost of each and names a control for each: the model an agent runs on, the spawn prompt it starts with, shutting it down, a spend cap, a time cap, and an output format that reports what a run cost. This guide collects those facts from Anthropic's own pages and ends with one real unattended job, Reveneau's own.
Claude Code usage limits: the two windows, the plans, and what to do when a limit stops you
A Claude Code usage limit is the allowance of work a Pro, Max, Team or Enterprise subscription includes, and it has two windows: a session window that resets every five hours and a weekly window that resets at a fixed time assigned to your account, by Anthropic's support articles read on 4 October 2026. Every Claude product on the subscription draws from the same allowance. When a window runs out, Claude Code shows the reset time and blocks requests until then, and you can wait, let Claude Code wait and continue on its own, turn on usage credits at standard API rates, upgrade, or pay per token with an API key. This guide explains each part, dates and attributes every fact, and keeps a log of the changes.
The Claude Code token benchmark: one task, nine settings, 45 runs
Reveneau's Claude Code token benchmark ran the same small coding task through Claude Code's non-interactive mode (claude -p) 45 times on 4 October 2026, five runs per setting across nine settings in four experiments, and recorded what Claude Code itself reported for each run: the dollar figure, new input tokens, output tokens, cache read tokens, cache write tokens and the number of turns. All 45 runs passed their test, and they cost $9.15 in total by Claude Code's own list-price figure on a subscription account, with Claude Code 2.1.118 and aliases resolved to claude-sonnet-4-6, claude-opus-4-7 and claude-haiku-4-5-20251001. This guide is the report: the method, the four results with their arithmetic, what the run cannot show, and the raw data.
Eval-driven development: how to prove AI-written code works
AI can write a feature in minutes. Proving that the feature does what you asked still takes real work, and that work is now the only thing that stops a fast team from releasing a broken product.
Why AI evals matter: the test that shows whether an AI product works
An AI product can give different output on different days and for different inputs, so a demo, a vendor's promise and a public benchmark score are weak evidence that it works for your customers. An AI eval is the stronger evidence: a written, repeatable test with an input, an expectation and a scoring rule. This guide is written for the people who approve budgets and releases. It explains what an eval is, what public records show about wrong AI answers and advertising claims, how to read a pass rate and use it to decide a release, what the work costs and who owns it, and what the texts of regulators say about testing.
LLM evals: how to measure whether an AI product works
LLM evals measure whether a product built on a large language model does its job. The method has a fixed order: learn the three kinds of eval, read real outputs and name the failure types, build a test set from real usage, size the set for the difference you need to detect, choose a score for each failure type, and keep measuring after release. A 90 percent pass rate on 100 cases has a 95 percent confidence interval of 84.1 to 95.9 percent, by the formula in Evan Miller's 2024 paper, so a small set cannot confirm a small improvement. Each step in this guide is tied to a named paper or vendor guide.
AI agent evals: how to test an agent that takes actions
An AI agent is a language model working in a loop: it picks a step, calls a tool such as a search or a database, reads the result and repeats until the task is done. Because it takes many steps and acts on real systems, a test of its final answer covers a small part of what can go wrong. This guide is the method. Grade the end state and the steps, test each tool call, repeat every case because one pass proves little, set limits on cost, time and steps, run the tests in a closed copy of the system, test for attacks, and turn production runs into new test cases.
AI benchmarks vs your own evals: how to read a model score
A benchmark score answers one question: how did this model do on that public test, under the settings chosen by whoever ran it. A buyer needs the answer to a different question: how will the model do on my work. This guide explains the benchmarks vendors quote, including SWE-bench, MMLU, GPQA, ARC-AGI and the agent tests, with each one's size and method from its own paper. It covers three ways a score misleads: the questions were in the training data, every model scores near the top, and the vendor chose the settings. It ends with a seven-step method for choosing a model on your own cases.
Faster evals with Jev: how we grade AI-written code
An eval suite for AI-written code has two kinds of check. Some have an exact answer, and code decides them. The rest used to need a language model to read the change and give an opinion, which was the slowest and least repeatable step in the suite. Reveneau moved that step to Jev, a decision model that answers fixed questions with probabilities instead of writing text, and on our own suite the run now finishes ten times faster than it did with a language-model grader. This guide explains how the grader works, how to design its questions, and where it stops being useful.
Jev and System One models: a plain guide
Jev is a hosted model from TypeSafe AI that answers fixed questions about a piece of text with calibrated probabilities instead of writing a reply. TypeSafe calls this a System One model. It returns a choice, a score or a yes/no probability in well under a second, and TypeSafe prices it at $0.042 per million input tokens with output free. This guide explains what that buys you, where the model is known to fail, and how to decide between Jev and a language model for each decision in your software.
Jev in production: guardrails, routing and classification
A decision model answers a fixed question about a piece of text and returns a probability instead of a paragraph. Inside an application that makes it useful in four places: routing a request to a handler, gating an action on how sure the model is, checking a value that code already extracted, and screening what goes into and comes out of a language model. This guide covers each of those with Jev, TypeSafe's System One model, pattern by pattern, with the vendor's own figures marked as the vendor's and the limits stated at the start.
Should your product use a decision model? A guide for founders and CTOs
A decision model answers a fixed question about a piece of text with a probability, in under half a second, at a price TypeSafe lists as $0.042 per million input tokens. It writes nothing. That makes it the right tool for a narrow set of product features, the wrong tool for most others, and a governance question either way. This guide is written for the person who approves the decision: how to sort features, how to run a two-week pilot, what to ask a vendor or partner, and how to keep the model auditable once it is in production.
How to deliver high-quality software with AI-native development
AI now writes code fast, so writing code is no longer the slow step. The slow steps are the work around it: deciding what to build, proving it works, releasing it safely, and running it afterwards. This guide describes how a team should organise all four, at the speed AI now makes possible.
Taking AI agents from prototype to production
An agent that works in a demo and an agent that works in production are different systems. The demo has to succeed once, with someone friendly at the keyboard. Production has to succeed repeatedly, for people who did not read the instructions, with real money and real data affected by each call.
Product discovery and UX design: a buyer's guide
Most failed software was built competently. The team delivered what was asked for, on a reasonable timeline, and it still did not matter, because what was asked for was never tested against what users actually needed. Discovery and design are how you buy that test before you pay for the build.
Building compliant software in financial services
Financial regulation is already written as rules. That is the part most teams miss. A rule that says a record must be recreatable after deletion is a specification, and a specification can be turned into a check that runs on every change. This guide covers how to do that, which obligations actually apply to your code, and why compliance gaps in a working app are so easy to miss.
Building compliant software in healthcare
HIPAA is unusual among regulations in that it tells you what to achieve and mostly declines to tell you how. That flexibility is why so many healthcare applications are non-compliant while holding a completed risk assessment: the rule left the decision to you, and nobody wrote the decision down. This guide covers what the Security Rule requires from software, and how to turn it into something a system enforces.
Building compliant software in real estate
Real estate software has a compliance problem the other regulated industries do not. In finance and healthcare, the rules mostly govern how you handle data. In housing, the rules govern the decision your software makes, which means the model's output is itself the regulated act. That changes what you have to build and what you have to test.
Building compliant software in insurance
Insurance regulation is state by state, which most teams treat as a reason to guess. It is really a reason to build the rule into configuration rather than into code, so the same application can satisfy Iowa and Ohio without a special case for each. This guide covers what the NAIC model laws actually require, which parts apply to your application, and why a working system so often looks compliant to its own team while missing the one thing an examiner asks for.
Building compliant software in legal technology
Legal technology carries a rule most other software does not: the product itself can end up practicing law, even when nobody intended it to. This guide covers where that boundary is, what the ethics rules that govern lawyers actually require of the tools they use, which e-discovery obligations affect your data model, and how to turn each of those into something a build can be tested against rather than a policy nobody rereads.
Software for compliance-heavy industries
Regulated software has always needed exhaustive verification and rarely received it, because writing the checks cost more than anyone would approve. That constraint is gone. This guide covers what that changes, the obligations that apply regardless of sector, and how to tell a partner who verifies their work from one who says they do.
Forward deployed engineering: the complete guide
A forward deployed engineer is a software engineer who works inside the customer's environment, on the customer's real data, and stays responsible until the system runs in production. The role exists because software that works in a demo and software that keeps working inside a real company are two different pieces of work, and somebody has to do the second one.
Top AI development companies in Silicon Valley
These are fourteen firms with an office in Silicon Valley or San Francisco that build AI software for other companies. Every fact on this page was checked on 20 October 2026, and every sentence says who the fact comes from: the firm itself, a news report, or a partner directory. No firm publishes a price, and no firm here gets a score.
Top forward deployed engineering firms in the US
These are fifteen firms with a US office, or a US company that works remotely, that sell forward deployed engineering: they place their own engineers inside a client's team to build software and keep it running. Every fact was checked on 20 October 2026, and each sentence says whether the fact comes from the firm itself, a news report, a filing or a partner directory. No firm publishes a price, and no firm gets a score.
Top forward deployed engineering firms in Europe
This guide lists seventeen firms that a company in Europe can ask for forward deployed engineering, meaning engineers from the firm placed inside the client's team to build software. Twelve describe this work on their own sites, four use the term only in job ads or as a job title, and one uses it only in its directory text. Twelve are based in Europe, three are US firms that name a European office, and two, Reveneau and Fluxon, are based in North America. Every fact was checked on 21 October 2026, and each sentence says whether it comes from the firm, a company register, a news report or a partner directory. No firm publishes a rate card, and no firm gets a score.
Top AI development companies in the US
These are nineteen firms with an office in the United States that build AI software for other companies. After Reveneau and Fluxon come five product studios, six firms that work mainly on AI and data, and six large consultancies with an AI practice. Every fact was checked on 21 October 2026, and each sentence says where it comes from: the firm itself, a filing with a regulator, a news report or a partner directory. One firm prints a price figure. No firm gets a score.
Top AI development companies in Europe
This guide lists twenty firms that build AI software for companies in Europe. Reveneau and Fluxon come first. Then come six product studios, six firms that work mainly on AI and data, and six large IT groups with an AI practice, and all eighteen of those name an office in Europe. Every fact was checked on 21 October 2026, and each sentence says where it comes from: the firm itself, a national company register, an annual report, an official statistics office or Anthropic's partner directory. One firm, Miquido, prints prices on the pages read. No firm gets a score.
Top AI agent development companies
An AI agent is a computer program that uses an AI model to finish a task in several steps by itself, such as checking an order, filling in a form or changing code, and that passes the work to a person when it should. This guide lists seventeen firms a company can hire to build agents for it. Reveneau and Fluxon come first. The other fifteen are all listed in Anthropic's partner directory and have offices in the United States, Canada or Europe. The sources were read on 20 and 21 October 2026, and each sentence says whose fact it is. No firm gets a score.
Best AI development partner for your situation
The right AI development partner depends on the situation your company is in. This guide covers five: a startup building its first AI product, a large company adding AI to the systems it already runs, a company in a regulated industry, a company whose earlier build failed, and a company that wants outside engineers inside its own team. Each situation has its own page. Reveneau is listed first and Fluxon second, then 34 other firms, each matched to a situation by the words on its own website. The firm pages that place each firm in a situation were read on 21 October 2026, and no firm gets a score.
Building your AI product after you raise: a founder's guide
This guide is for a founder in the first year after a seed or Series A round who is building a product that uses AI. Carta's report for the first quarter of 2026 puts the median time from seed to Series A at two years, so the money has to pay for a product that earns the next round. The guide covers how long the money lasts, what it pays for, what the next investors look at, whether to use a model another company runs or train your own, what each user costs to serve, provider data terms, code ownership, security reviews and the AI claims regulators have acted on.
AI in banking: how a bank uses AI under the rules it answers to
This guide is for a bank executive who has to decide how the bank uses AI and which rules apply. Several of those rules changed in 2025 and 2026. On 17 April 2026 the US banking agencies replaced their 2011 model risk guidance with new guidance that does not bind banks and leaves generative and agentic AI out of scope. The Consumer Financial Protection Bureau withdrew its two circulars on AI and credit denials on 12 May 2025. The EU moved its credit scoring date to 2 December 2027. Each rule here carries its status as of 26 October 2026.
Enterprise AI adoption: how a large company catches up on AI
This guide is for an executive at a large company that wants to catch up on artificial intelligence (AI). The US Census Bureau's newest survey, collected 5 to 18 October 2026, found that 43.6% of US businesses with 250 or more employees had used AI in the previous two weeks. That figure moved by up to 7.5 points between neighbouring survey periods. In the European Union, 55.03% of large enterprises used AI in 2025. The guide covers the figures, staff training, the first work to choose, governance, security, data, older systems and how to measure returns, with each rule's status as of 26 October 2026.