Charity has a trio of lovely AI posts up, “AI enthusiasts are in a race against time, AI skeptics are in a race against entropy”, “AI demands more engineering discipline. Not less.”, and “Make AI boring again.” On the one hand, they are perfect trilogy from someone whose life’s work has been making product development and management just fundamentally better (see, e.g. the 2nd edition of Observability Engineering just released!). On the other hand, they perfectly capture the unfortunate reality that AI — even more than news in general — is now the easiest topic for outrage farming, so in our attention reinforcement world everyone from politicians to CEOs are playing to the mob rather than having serious discussions about AI. (Lots of otherwise smart people still don’t seem to actually understand what happened as deep learning took over ranking in the 2010s.)

However, on the gripping hand, the real question right now is whether you are an engineer, tech manager, or exec at most any company: what the hell do I do?.

(No, it’s not banning AI at your company. Your teams will be using it anyway, likely with cheap consumer accounts so when employees cut-and-paste your customer data into the web chat interface, the provider will definitely use it for training.)

In the spirit of “blogging can just be stating the obvious”, my do’s and don’ts for AI in mid-2026.

Recognize that AI is generating strong emotions with your people, so even seemingly simple discussions will be fraught.

What to do as an exec?

Make affirmative decisions about AI use and measurement. Require AI proponents to show their receipts.
Be clear about AI norms. Don’t tolerate slop grenades.
Model the business impact of a competitor applying AI successfully. Spend accordingly.
AI will compound existing problems in processes and communication. Don’t blame AI, use the new information to fix the problems.
Actually use AI yourself in a work context. This is too meaningful a moment to be printing out emails.
Don’t create a token leader board
Ensure AI is multiplayer, not isolating
Don’t use AI as an excuse to fire people. Don’t stop hiring junior people.
Whatever your decision, identify the laggards. Acknowledge their concerns, then take away their megaphones.

What to do as a tech manager

Figure out how many measurable, time-gated experiments you can afford. Get them rolling loudly and transparently. Learn from the failures.
Don’t bet on one model or model provider, no matter how out in front they seem.
Ensure your teams aren’t building on sand or jello. Do the work to fix your foundations.
Build loops with meaningful feedback. Agents without closed loops are just guessing.
Find a way for agents to improve your product with minimal oversight. Then find the next way.
Understand risk like never before. Embrace that code review is about shared liability as much as shared knowledge.
Recognize the challenges inherent to operating in highly ambiguous, unknown-unknown states. Allocate money, time, and attention to hedges and leapfrogs.
Pick a date to become a tech island. Hit it.

What to do as an engineer

Decide what parts of engineering you truly love and choose accordingly. If you love artisanal code, what you type better be extremely specialized or have no competition.
Don’t go halfway. AI-as-better-tab-completion was briefly interesting in 2025.
Don’t burn out. Agents don’t get tired and will consume all the time and attention you can give them.
Share what you’ve been learning. Including the failures.
Get excited about the foundations. Data structures, security models, and foundations that agents can bounce off of safely are the new table stakes.
Don’t get sucked into the outrage on either side. What pundits say about AI efficacy doesn’t matter. Does AI make you, your teams, your colleagues, your products better? If not, what would need to change? What do you need to do in order for that improvement to be under your control?