DevHawk

DevHawk is the AI PM tool that boosts productivity for outsourced teams by following up with developers, holding them accountable, and keeping projects moving...so you don’t have to.

Why do folks with a technical background still outperform when using agents to build software? And how does this change as models improve? Engineers know when to pick option 3! Right now the best models seem to get things right 90% of the time - my rough guestimate. Perhaps 95% of the time in the good case. So if you hit 1 every time, you're letting error creep in every 10-20 executions. Sounds small, but it begins to accrete quickly. With our software factory at DevHawk, we see the difference among clients - and the most dangerous of all are the ones who think they are "developers" now. So they build complex features, merrily hitting 1 or just going full auto. The PR comes up for review. It's a bloodbath lol, maybe just a straight rejection. My take: over the coming year models will continue to improve, but they won't touch the magic line of perfection. And so having engineers around to help redirect them will be a good idea for some time to come!

A biz truth in this: software is becoming a commodity, but relationships are not. This is why we see conferences and events booming, even as AI is the topic everywhere. This is why DevHawk moved away from fractional remote first, and is hiring in office in ATL. This is why (for the moment?) human voices still can stand out in the sea of AI slop. Does that change as AI advances, as humanoid robots become like those in BladeRunner? Perhaps someday - but today and for the foreseeable future, human relationships are still, well, human. Hope to see more of y'all IRL sometime!

A healthcare platform asked me a simple question: can you build a tool that lets physicians walk patients through treatment options live? As I dug in, I realized they were actually conflating two different products they needed: - The clinical specialists would be power users, logging in daily, building protocol comparisons and reviewing expected outcomes. Depth required: edge cases and complex multi-condition scenarios - The primary care docs needed to pull up a clean visual comparison during a patient visit: "Here's what happens if we stay the course, here's what a medication change looks like, here's the procedure path" and have a patient understand it in real time. One of our engineers showed a mockup on the call. The clinical director's first reaction was: this is cleaner than what we're using today, but will it handle the complex cases our specialists need? The simplified view that works in a patient conversation can feel inadequate for the experts, and the expert view that captures everything is too much to use with patients. We handled both - with progressive functionality and RBAC (role-based access control). But AI's tendency is to either dramatically overbuild (or underbuild) by default - we've still got to bring the nuance to handle both needs.

I don't think the old SaaS multiples are going to come back. So what's the move for those of us still in/around software? First, let me frame the dilemma: - The TCO (total cost of ownership) of a small software product has fallen from 500k/year to 100k per year for a wide range of use cases. - You used to need a team of multiple folks - now arguably you can get away with less than one full salary. - This still sounds expensive compared to SaaS, but if that product is customized to the precise needs of one firm, it can replace a half-dozen old SaaS products. B2C might be more defensible today, because no one is going to build software to replace a single $10 subscription. But as the models continue to improve, what's the endgame for B2B software, even AI native software? I think there's a new market structure coming, with roughly the following choices: 1. SaaS: Prices will fall and discounts will rise 2. Build option: more and more teams will build on their own 3. AI Native Players: this is just SaaS 2.0 - why can't your customer build this? Long term they will be subject to the same forces as SaaS but with more embedded token costs and lower margins New Options: 4. Custom SaaS: Vendors will build custom SaaS with N=1, hyperspecialized hosted solutions. Cheaper than building, better than buying? 5. Software Factories? Provide the client with the hard parts (auth, security, integrations) and guardrails (agents watching for security holes etc), and let them build whatever features they want atop it. (This is DevHawk) It's Mad Max right now in the Post-SaaSpocalypse world - What other options do you think will emerge?

Everyone keeps saying token prices are about to rise... who you gonna believe, the naysayers or your lyin' eyes??? I mean, OpenAI slashing prices by 80!? Since the price change their usage has apparently spiked on OpenRouter as well, proving that price does matter. Both Anthropic and OpenAI are locked in an intense fight against: 1. Cheaper Chinese model inference purchased directly or from any number of model hosting providers 2. The ability of clients to run those Chinese models locally, on smaller and smaller hardware - see the latest Qwen 3.8 drop OpenAI is directly fighting back with price cuts. Anthropic views themselves as the Mercedes of the category - but they are quietly cutting prices too. How? Anyone else notice they've stopped all the threats about kicking you off Claude Max if you run your agents on it? The real story here is - AI is here to stay and is going to keep getting cheaper, faster, and more ubiquitous. It's on you to figure out how to use it - or to ask someone who knows!