Both Columns, Honestly

Our position on AI, written for arguing with. Both columns are here, honestly. The part that matters is the middle, and the middle is a list of decisions.

Most weeks, inside a small consulting or professional services firm, the same argument happens again. One partner watched a research task shrink from a day to an hour and wants the technology everywhere. Another watched it invent a source with a straight face and wants it nowhere. The argument is loud, rarely settled, and almost never written down.

This piece is our attempt to write it down. We think AI is one of the biggest technology shifts of our lifetime, and we use it a lot. We also reject blind optimism and blind opposition. Neither extreme helps you run a firm. Here are both columns, honestly, then the part that matters most.

The whole page at a glance

Most takes hand you only their favorite row. Here is each one in full.

The good column

In a small firm, the real gain shows up as time. Research that used to eat a day can come back in an hour, with sources to check. First drafts stop being a blank-page problem. Data nobody had time to analyze gets analyzed. Administrative work stops spilling into evenings.

None of that has to replace anyone. Removing repetitive, low-value work makes room for the parts clients are actually paying for: judgment, relationships, creativity, problem solving. The same capability is being pointed at scientific discovery, healthcare, and access to knowledge, and none of it should be waved away because risks exist. Progress has always brought challenges. Challenges have never been a good reason to stop.

But notice something. Nothing in this column separates one firm from another. Every vendor homepage says some version of it, and so do we, because it is true. That is exactly the problem with stopping here: a column everyone shares cannot be anyone's policy.

All of it true. Only half the page.

The bad column

This column is real too. It needs no adjectives; the record is enough.

Fabrication. These systems make things up, fluently. We have experienced it firsthand: fabricated information, wrong assumptions, outputs that looked right and were not. In October 2025, Deloitte agreed to a partial refund to the Australian government on a AU$440,000 report after fabricated citations were found in it (Associated Press, October 7, 2025). Left unchecked, that failure mode carries reputational, legal, and financial risk for a firm of any size.

Environment. The infrastructure behind these tools is physical. The IEA projects that data center electricity consumption will roughly double to around 945 TWh by 2030, just under 3 percent of global electricity use (IEA, Energy and AI, 2025). The same IEA analysis estimates data centers already consume around 560 billion liters of water a year, a figure that could roughly double to around 1,200 billion liters by 2030. Those numbers deserve scrutiny rather than panic or dismissal. Any firm adopting the technology becomes, in a small way, part of that demand curve.

The harder-to-number questions. Workforce disruption, privacy, misinformation, and intellectual property all raise questions that deserve serious discussion, not a shrug.

The row almost nobody names

There is a third row on the page, and most positions skip it: cannot know yet. Nobody knows the long-run workforce effects. Nobody knows whether the capability curve keeps climbing or flattens. Nobody knows what regulation will ask of a thirty-person firm in three years. Both tribes write confidently in this row anyway. The honest entry is empty, and your firm has to decide things now without it.

The tell

Both poles are popular for the same reason: each one deletes your to-do list. Go all in, or ban it all. Either way the deciding is over and nobody has to do the work.

Here is a quick test for whether a firm has done the work. Read its current AI position and count the nouns from its own workflow. Proposal drafts. Client data. The QA pass. The Tuesday status meeting. A stance contains none of them; a policy is full of them. If yours has none, you have a tribe, not a policy.

The list of decisions

Here is the middle, written out: the nine decisions a firm of five to thirty people actually has to make, each with realistic options. Bring it to your next partner meeting. It is the agenda.

1. What work may it touch? Options: research and internal drafts only. Anything, as long as a person reviews it before it leaves the firm. Nothing client-facing. Or decided engagement by engagement.

2. Who reviews the output, and how hard? Options: the person who used the tool. A second reader for anything client-facing. Owner review for high-stakes deliverables.

3. What gets verified before it ships? Options: every citation, number, name, and quote checked against a source. Full checks for external documents only. Spot checks.

4. What is off-limits entirely? Options: final recommendations. Anything carrying a signature. Personnel matters. Or nothing, if the review rules hold.

5. What client data may enter a tool? Options: none. Anonymized only. Only with written consent. Only tools whose data terms your firm has actually read.

6. Which tools are approved, and who approves new ones? Options: a short list with one owner. Anyone proposes, one person decides. Decide personal accounts here too.

7. Whether and how to tell clients? Options: a standing note on deliverables. A clause in the engagement letter. An honest answer when asked. Each carries a different cost.

8. Who is accountable when it is wrong? Options: the person who shipped it, as with any other work. Or shared with the reviewer. The option that does not exist: the tool.

9. When will you revisit all of this? Options: quarterly. Whenever a tool changes materially. After the first failure. The only wrong answer is never.

An afternoon with this list can produce a first draft your firm can actually follow. Our own conversation ended where it had to: take the technology seriously, put people and the environment first, support the slow work of regulation without waiting for it, and keep a human accountable for everything that leaves the building.

The poles will keep winning arguments, because each pole ends the deciding. Your firm does not live at either one. This was never all-or-nothing. A position you can only agree with is a tribe; a position with nine decisions in it is an agenda. The middle is a list of decisions.

Continue the series

This is part 1 of a 5-part series on Both Columns, Honestly. The full arc:

• Monday: Both Columns, Honestly (this post)
• Tuesday: The Side Spreadsheet Is Back (coming Tuesday)
• Wednesday: The Deliverable Was Always the Receipt (coming Wednesday)
• Thursday: The Research Says This Footer Should Cost Us Trust. We Use It Anyway. (coming Thursday)
• Friday: You Are Less Behind Than Your Feed Says (coming Friday)

Content and Accuracy Disclaimer

This article was drafted with AI assistance and reviewed by the Altvina team. We rigorously fact-check all content to ensure reliability.

Should you notice any inaccuracies or outdated information, please contact us so we can correct it. Your feedback helps us maintain high standards of accuracy and transparency.