Thoughts on technology leadership, efficient engineering, and building scalable organizations.
We've held the founding-level seats where the expensive calls get made, in technology, teams, and business, and made them before they got expensive.
AI runs inside our delivery as an operating model, not a slide. The judgment to direct it is the whole point.
Flat monthly retainers. You measure us on what we ship and what we prevent, never a timesheet.
How your systems fit, where they bend, and the path forward, written down and kept current, so speed never outruns understanding.
Your team says the AI got worse this month. Without a dated record of your own work, that argument has no floor and no end.
Your team says the AI got worse and nobody can settle it. What to record in your own workflow now, so the very next complaint has a dated before to check.
By The Bushido Collective
An hourly meter prices your advisor's time. It also teaches you to batch the questions whose value you can't judge until after you've asked one.
An hourly meter prices your advisor's time and quietly teaches you to batch the small questions. Two threads this week price that habit from both sides.
By The Bushido Collective
Before you automate it, find out whether the person running it can still tell you what they do.
The documented process is the happy path. The exceptions live in one person's reflexes, which is why interviews, binders, and AI pilots keep missing them.
By The Bushido Collective
Broadcom stopped publishing the library third-party tools use to read VMware disks. The thing that gets your data out was never yours to begin with.
The library migration tools use to read VMware disks stopped being downloadable. How to test whether your exit path depends on the vendor you would leave.
By The Bushido Collective
An expired domain, bought for five euros, logged 209,205 lookups carrying the numbers other people's phone systems were about to dial
A five-euro expired domain logged 209,205 dialed-number lookups from working phone systems. Your systems call addresses nobody owns, and nothing screams.
By The Bushido Collective
A chatbot can review your proposal through your client's suspicions. Give it the facts, and leave room for it to find a real mistake.
When a client asks AI to review your recommendation, the prompt can change the verdict. Make the reasoning checkable, including the case for a cheaper option.
By The Bushido Collective
The gain on one person's calendar can become a cost on another's.
AI can speed up a draft while creating work for the person who checks it. How to measure uneven gains and choose between review, coaching, and narrower use.
By The Bushido Collective
Your software supplier can bundle AI into the renewal. It can't decide which work is worth changing.
AI comes bundled into software you already buy. A renewal price hike and a support study show how to judge the work it saves, including the cost of checking.
By The Bushido Collective
Incident command has to reach the agents acting on your engineers' behalf.
Two approved agent actions can undo each other. How incident command can govern queued changes and track in-flight work without stopping parallel investigation.
By The Bushido Collective
A good demo shows a selected case. Before you buy, specify the work the tool must finish and what should happen when it can't.
Evaluate an AI demo without reading code: specify the finished job, test failure and handoff, then compare the trial with the work your staff does today.
By The Bushido Collective
Owning the code doesn't settle who has the experience to build and run it.
Separate AI delivery gaps from the build-versus-buy debate: test the failing workflow, weigh internal capacity, and make an outside builder earn the same trust.
By The Bushido Collective
A ranked backlog can be correct on paper and still leave you with the hardest part of the decision.
AI can speed up management writing without resolving the decision behind it. Separate missing context from the work of earning trust and following through.
By The Bushido Collective
A polished codebase leaves a question open: who can answer for what the model built?
AI can write the code, tests, and README. A payment retry shows how acquisition diligence can test whether the team understands and can operate what it owns.
By The Bushido Collective
You measured attendance and called it adoption.
AI usage grades can reward compliance while work stalls. A real mandate and a measured support rollout show what to check before calling adoption a success.
By The Bushido Collective
The offer you'd approve for a replacement belongs in the conversation about the engineer who stayed.
When hiring pay outruns incumbent pay, the real decision includes peer corrections, replacement costs, and who can approve spending before an engineer leaves.
By The Bushido Collective
A new manager needs time you've already committed to their old job.
A promotion can double-count one person's time. Separate a poor management appointment from the workload and development the company left out of its plan.
By The Bushido Collective
A management title changes who does the work. Your release plan needs to change with it.
Promoting a great engineer changes what the team can deliver. Account for the handoff, test the AI assumptions, and give a new manager room to learn the job.
By The Bushido Collective
Automating the work can also remove the practice. Which decisions are your senior engineers still exercising?
AI can remove routine work and the practice behind sound decisions. How engineering leaders distinguish wasted effort from work that keeps judgment in use.
By The Bushido Collective
Phoneware's rebuild connected product answers to human support. A site its own team could maintain was the harder promise.
Phoneware's rebuild connected static pages and an AI concierge to human support. The harder question was whether its team could maintain the site unaided.
By The Bushido Collective
A code freeze didn't stop a database deletion. An approval button won't help either if nobody can inspect what it releases.
AI oversight depends on which actions an agent can reach, who authorizes them, and whether that person can check the work before a draft becomes a commitment.
By The Bushido Collective
A routine-sounding call can ask for a decision your business hasn't made.
An AI receptionist can explain a warranty without approving a repair. Map what each promise requires, who can decide, and what happens when no one answers.
By The Bushido Collective
A senior title can conceal missing judgment. A stalled promotion can conceal someone already supplying it. The remedy depends on which problem you have.
Before hiring a staff engineer, use a cross-team rollout to distinguish missing technical experience from missing authority, time, or recognition in the team.
By The Bushido Collective
More developers can clear a backlog or give your busiest engineer more decisions. Check which you're buying.
More contractors can add useful capacity or overload the people who approve their work. Learn how to judge a staffing proposal, including its claims about AI.
By The Bushido Collective
A prototype can answer a useful question without earning a permanent place on the roadmap.
Cheap demos can reveal what deserves investment or turn every idea into an obligation. How to keep experimentation broad without letting commitments multiply.
By The Bushido Collective
AI can answer a question without letting a teammate know why you're asking.
AI can replace a colleague's answer without replacing their awareness. Where that gap matters, and how to reward shared understanding without more meetings.
By The Bushido Collective
When nobody can explain an AI-built system, recovering the old design may be the wrong job.
An AI-built codebase can outlive the reasoning behind it. How to separate recoverable design from leftover code, and take responsibility for what runs now.
By The Bushido Collective
Making preventive work visible without turning it into another rescue story.
Preventive work needs evidence beyond a clean quarter. Show the risk, the cheaper alternative, and the test result without inventing disasters or savings.
By The Bushido Collective
A smaller system can be worth a worse metric. The hard part is knowing which loss the business can accept.
Knowing what not to build takes more than skepticism. Segment's cache tradeoff shows how to make a refusal accountable to the problem it is meant to solve.
By The Bushido Collective
A flat bill buys room to learn which agent workflows are worth repeating.
A flat AI bill can hide the cost of an agent workflow. Learn to separate useful exploration from habit before your team depends on more usage than it covers.
By The Bushido Collective
An activity report can mistake missing data for missing work, then leave an engineer to explain the difference.
Before using an activity dashboard to judge engineers or AI spending, check whether it can distinguish an empty record from work it never managed to observe.
By The Bushido Collective
A growing PR queue can mean too few reviewers, or too much understanding left for them to supply.
A growing pull-request queue can hide distinct problems. Distinguish missing reviewers from missing evidence, and check whether a remedy saves work or moves it.
By The Bushido Collective
The handoff is still unfinished when the next engineer has the code but can't recover the reasoning.
A handoff needs more than code and a walkthrough. Test whether the next owner can recover the reasoning, handle a failed change, and know when to stop safely.
By The Bushido Collective
An AI-built tool needs permission to do its job. How much more did you give it?
An AI-built tool needs more than a working demo. Learn how to limit its access, test uncertain writes, and decide when a reviewed import is the safer option.
By The Bushido Collective
What human approval contributes, and when it limits the return on your AI investment.
AI can speed up drafting without speeding up delivery. A practical way to test what human review contributes before deciding which work can run without it.
By The Bushido Collective
Faster coding can leave delivery unchanged. The budget decision depends on where the saved effort goes.
AI coding tools can save time while delivery stays flat. Trace the review burden, find the real constraint, and decide which spending has earned its place.
By The Bushido Collective
Let the next owner do the explaining
A runbook doesn't prove someone else can run the service. A practical way to transfer judgment and test the handoff while the expert is still there to help.
By The Bushido Collective
A request explains where a feature came from. Someone still has to make the case for building it.
A request can outlive its reason. Recheck the problem, compare alternatives, and distinguish an old roadmap idea from a promise the business needs to honor.
By The Bushido Collective
Follow an access-review claim from the source records to the auditor's test before promising a buyer a report.
How to assess a fast SOC 2 offer, trace an access-review claim to real evidence, and decide with your buyer whether the report provides the assurance they need.
By The Bushido Collective
Your team's assistants can agree with each engineer while the engineers disagree with each other.
When AI-backed answers collide in design review, compare the assumptions behind them. A practical way to make technical judgment visible and open to challenge.
By The Bushido Collective
A headcount plan can balance while the next generation of system owners goes unfunded.
AI can help ship the next release without developing the next system owner. How to check learning, fund mentoring, and judge when hiring experience makes sense.
By The Bushido Collective
Before another senior hire, find out which work actually left when the junior roles did
A senior-heavy team can still lack capacity. Separate expert maintenance, work without an owner, and the tasks that teach before changing your hiring plan.
By The Bushido Collective
When AI lowers the cost of building, a useful experiment still needs a question it can answer and a decision someone will own.
Cheaper code can remove a crude filter on ideas. Keep the scrutiny without blocking useful experiments: budget for evidence, support and a way to stop.
By The Bushido Collective
Before blaming the approval process, count the time spent checking and correcting the AI's work.
AI adoption can rise while delivery stays flat. Separate task savings from waiting time, then test whether the tool or the approval process needs to change.
By The Bushido Collective
A stalled launch needs a diagnosis before it needs another feature.
A working product can stall before a customer ever uses it. Trace the path from first contact to payment to find which problem your next feature would solve.
By The Bushido Collective
A polished submission leaves a hiring question open: who can tell when the code is wrong?
AI can pass a coding test without revealing a candidate's judgment. Use the follow-up to see how they verify code, investigate failures, and change course.
By The Bushido Collective
A missing owner, an overloaded leader and the wrong person in the job demand different responses.
Before adding a Chief AI Officer, separate missing authority from missing capacity or skill. A practical case for one technology leader, and when to hire two.
By The Bushido Collective
How to tell whether a fractional CTO is earning trust, getting blocked, or failing to lead.
Evaluate a fractional CTO's first 90 days by following a recommendation from diagnosis to action, separating weak leadership from constraints you need to own.
By The Bushido Collective
A working payment flow leaves a harder question: can you explain what happens when part of it fails?
AI can help you ship before you understand the system. Trace a payment failure, see where retry protection ends, and learn how to check the repair for yourself.
By The Bushido Collective
Faster tasks leave a management decision behind: which handoffs still earn their place?
AI can speed up a task without changing the result. Follow a refund through review and exceptions to test which handoffs to keep, and where savings come from.
By The Bushido Collective
How a search for certainty can leave engineers caught between two founders
How founder-CTO distrust turns second opinions into competing commitments, and how to examine the breakdown without forcing engineers to choose between you.
By The Bushido Collective
Buying the tools is easier than reopening the rules. Some rules can go; others still protect your customers and money.
Uber and Klarna's AI accounts sharpen a board question: which work improved, at what cost, and which customer protections still have to hold after the rollout?
By The Bushido Collective
The people an AI plan proposes to remove may be the people it needs to work
An AI staffing plan can cut the people its success depends on. How to test whether gains survive the review, escalation, and staffing the service still needs.
By The Bushido Collective
A swallowed database error shows the limits of automated approval
A swallowed database error shows where Han's test hooks help, where they stop, and how to check if a passing test distinguishes failure from an empty result.
By The Bushido Collective
A missed date can't tell you whether an assumption broke or execution slipped
A missed milestone can hide a broken assumption or unfinished work. Make the assumptions testable so a Q1 review can distinguish learning from missed delivery.
By The Bushido Collective
Which part of the next feature are you paying for again?
Technical debt hides inside feature work. Trace the recurring effort, separate it from repair, and decide which changes can reduce the cost of shipping safely.
By The Bushido Collective
A hiring plan counts people. It also needs to account for work a technical decision could remove.
A slipping roadmap can hide work that a better technical decision would remove. How to tell when you need more engineers and when to use the team differently.
By The Bushido Collective
A feature can work exactly as designed and still deserve to be cut.
Work backward from a roadmap item to the business bet behind it, the evidence that would change it, and the named decision that removes work from the backlog.
By The Bushido Collective
The login works. Who can read the invoice?
A working login can hide a missing permission check. Learn how to test what an AI-built app must refuse before trusting it with your customers' private data.
By The Bushido Collective
Experienced engineers can get slower with AI. The useful question is what happens after the first draft.
How to review three AI-generated changes that pass CI, what the Copilot and METR studies measured, and how to tell saved effort from work moved to review.
By The Bushido Collective
A foundation should earn its keep before you know whether the company will scale.
Early foundations earn their keep by removing work from the next change. A distinction between useful conventions and bets on growth that may never come.
By The Bushido Collective
A lower fee helps only if the rest of the job has an owner. Compare the whole arrangement, including the people you already employ.
Compare fractional and full-time CTOs by decision authority, access to company context, and the cash and management capacity each arrangement actually needs.
By The Bushido Collective
An advisor's willingness to take equity deserves a closer look at what each side is giving up.
Advisor equity can support a lasting partnership. Judge the fee discount, the work behind the grant, and the reasons a good advisor might still say no.
By The Bushido Collective
Smaller teams can inherit the same bottleneck. Follow a delayed change before drawing the next boundary.
More engineers can mean more work waiting on the same people. Trace a delayed change to tell missing context from a dependency that needs a different remedy.
By The Bushido Collective
A fixed fee can reward efficiency while still discouraging the best advice.
Hourly billing can put useful advice in conflict with revenue. Fixed fees change that incentive, but scope, quality, and the option to stop still need a price.
By The Bushido Collective
A thirty-minute conversation about how your business runs, then a short written report: where AI actually pays off for you, and what it's roughly worth. No quizzes, no gimmicks, yours to take to anyone.
See how it worksShort, no-fluff strategy for technical leaders, every week or two. The thinking we bring to our engagements, no spam, unsubscribe any time.
You're in. Check your inbox to confirm, and for what we sent.
